Merge branch 'dev' into test/oversized-test-split-plan-3983

Co-authored-by: Alexandre Teixeira <111787685+alteixeira20@users.noreply.github.com>

.env.example

@@ -56,6 +56,13 @@ SEARXNG_INSTANCE=http://localhost:8080
 # SQLite database path (default: sqlite:///./data/app.db)
 # DATABASE_URL=sqlite:///./data/app.db
 
+# ============================================================
+# Data directory
+# ============================================================
+# Move everything that lives under data/ - settings, sessions, database, auth,
+# cache, uploads, etc. - to another path:
+# ODYSSEUS_DATA_DIR=C:\path\to\dir
+
 # ============================================================
 # Auth & Security
 # ============================================================
@@ -112,6 +119,9 @@ SEARXNG_INSTANCE=http://localhost:8080
 # Default: http://{LLM_HOST}:11434/v1/embeddings (ollama)
 # EMBEDDING_URL=http://localhost:11434/v1/embeddings
 
+# Embedding API key (if there's one)
+# EMBEDDING_API_KEY=embedding_api_key_here
+
 # Embedding model name (must be available at the endpoint above)
 # EMBEDDING_MODEL=all-minilm:l6-v2
 
@@ -144,6 +154,21 @@ SEARXNG_INSTANCE=http://localhost:8080
 # if you intentionally want scheduled scripts to run remotely.
 # ODYSSEUS_SCRIPT_HOST=localhost
 
+# Chat / agent attachment size cap in bytes (default: 10 MB).
+# Raise this for local installs that need larger PDFs or text documents.
+# Example: 52428800 = 50 MB.
+# ODYSSEUS_CHAT_UPLOAD_MAX_BYTES=10485760
+
+# Other per-feature upload size caps in bytes. All are validated and optional;
+# defaults shown. An invalid value (non-integer or < 1) fails fast at startup.
+# ODYSSEUS_GALLERY_UPLOAD_MAX_BYTES=104857600            # gallery image upload (100 MB)
+# ODYSSEUS_GALLERY_TRANSFORM_UPLOAD_MAX_BYTES=26214400   # gallery transform input (25 MB)
+# ODYSSEUS_MEMORY_IMPORT_MAX_BYTES=10485760              # memory import file (10 MB)
+# ODYSSEUS_PERSONAL_UPLOAD_MAX_BYTES=26214400            # personal document upload (25 MB)
+# ODYSSEUS_EMAIL_COMPOSE_UPLOAD_MAX_BYTES=26214400       # email compose attachment (25 MB)
+# ODYSSEUS_STT_MAX_AUDIO_BYTES=26214400                  # speech-to-text audio (25 MB)
+# ODYSSEUS_ICS_MAX_BYTES=10485760                        # calendar .ics import (10 MB)
+
 # ============================================================
 # GPU support (Docker Compose)
 # ============================================================

.github/CODEOWNERS

@@ -0,0 +1,8 @@
+# Code owners.
+#
+# Every file is owned by the maintainer, so that when branch protection has
+# "Require review from Code Owners" turned on, no pull request can be merged
+# without the maintainer's review. This is the human gate that backs up the
+# automated security checks. See docs/security-ci.md for how to turn it on.
+
+*       @pewdiepie-archdaemon

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -23,7 +23,7 @@ body:
           required: true
         - label: This is **not** a security vulnerability. (Vulnerabilities go to [GitHub Security Advisories](https://github.com/pewdiepie-archdaemon/odysseus/security/advisories/new) — see [SECURITY.md](https://github.com/pewdiepie-archdaemon/odysseus/blob/main/SECURITY.md).)
           required: true
-        - label: I am running the latest code from `main`.
+        - label: I am running the latest code from the `dev` branch (the default branch you get on clone, where fixes land first) and the bug still reproduces there. Please `git pull` the latest `dev` before filing.
           required: true
 
   - type: dropdown

.github/dependabot.yml

@@ -0,0 +1,48 @@
+# Dependabot keeps dependencies and pinned action versions current.
+#
+# Why this matters for security: every workflow in this repo pins its GitHub
+# Actions to an exact commit (a SHA), which is safe but freezes them in time.
+# Dependabot opens a small, reviewable pull request whenever a newer version
+# exists -- for Python packages, npm packages, the Docker base image, and the
+# pinned Actions themselves -- so staying patched does not require manual work.
+# Updates are grouped so a week's bumps arrive as one PR per ecosystem, not a
+# flood of separate ones.
+
+version: 2
+updates:
+  # Python dependencies (requirements.txt + requirements-optional.txt).
+  - package-ecosystem: pip
+    directory: "/"
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 5
+    groups:
+      python:
+        patterns: ["*"]
+
+  # Frontend / tooling npm packages (package.json).
+  - package-ecosystem: npm
+    directory: "/"
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 5
+    groups:
+      npm:
+        patterns: ["*"]
+
+  # The pinned action SHAs used across .github/workflows.
+  - package-ecosystem: github-actions
+    directory: "/"
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 5
+    groups:
+      actions:
+        patterns: ["*"]
+
+  # The Docker base image in the Dockerfile.
+  - package-ecosystem: docker
+    directory: "/"
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 5

.github/scripts/check-pr-description.js

@@ -103,14 +103,21 @@ module.exports = async ({ github, context, core }) => {
 
   async function swapLabel(num, add, remove) {
     if (await labelExists(add)) {
-      await github.rest.issues.addLabels({ owner, repo, issue_number: num, labels: [add] });
+      try {
+        await github.rest.issues.addLabels({ owner, repo, issue_number: num, labels: [add] });
+      } catch (e) {
+        // Fail soft on a token that can't write labels so a label permission
+        // problem never masks the actual description verdict.
+        if (e.status !== 403) throw e;
+        core.warning(`Could not add "${add}" — token lacks label write here; skipping.`);
+      }
     } else {
       core.warning(`Label "${add}" does not exist in the repo — skipping. Create it once to enable labelling.`);
     }
     try {
       await github.rest.issues.removeLabel({ owner, repo, issue_number: num, name: remove });
     } catch (e) {
-      if (e.status !== 404 && e.status !== 410) throw e;
+      if (e.status !== 404 && e.status !== 410 && e.status !== 403) throw e;
     }
   }
 

.github/workflows/ci.yml

@@ -20,6 +20,8 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+        with:
+          persist-credentials: false
       - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5
         with:
           python-version: "3.11"
@@ -31,6 +33,8 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+        with:
+          persist-credentials: false
       - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020  # v4
         with:
           node-version: "20"
@@ -51,10 +55,40 @@ jobs:
     continue-on-error: true
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      # Detect whether this PR only touches documentation files.
+      # If so, skip the expensive pytest run while still reporting a passing check.
+      - name: Check for docs-only changes
+        id: docs-check
+        run: |
+          if [ "${{ github.event_name }}" = "pull_request" ]; then
+            BASE="${{ github.event.pull_request.base.sha }}"
+            HEAD="${{ github.event.pull_request.head.sha }}"
+          else
+            BASE="${{ github.event.before }}"
+            HEAD="${{ github.sha }}"
+          fi
+          # List all changed files; if every file matches docs/markdown patterns, skip pytest.
+          changed=$(git diff --name-only "$BASE" "$HEAD" 2>/dev/null || git diff --name-only HEAD~1 HEAD)
+          non_docs=$(echo "$changed" | grep -Ev '^(docs/|.*\.md$|\.github/[^/]+\.md$)' || true)
+          if [ -z "$non_docs" ]; then
+            echo "docs_only=true" >> "$GITHUB_OUTPUT"
+            echo "Docs-only change detected — skipping pytest."
+          else
+            echo "docs_only=false" >> "$GITHUB_OUTPUT"
+          fi
+
       - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5
+        if: steps.docs-check.outputs.docs_only != 'true'
         with:
           python-version: "3.11"
           cache: pip
       - run: pip install -r requirements.txt
+        if: steps.docs-check.outputs.docs_only != 'true'
       - run: mkdir -p data  # sqlite DB lives at ./data/app.db
+        if: steps.docs-check.outputs.docs_only != 'true'
       - run: python -m pytest -q
+        if: steps.docs-check.outputs.docs_only != 'true'

.github/workflows/codeql.yml

@@ -0,0 +1,61 @@
+# CodeQL code scanning
+#
+# Purpose: GitHub's own static analysis engine reads the application source
+# (Python backend + the JavaScript frontend) and looks for real
+# vulnerabilities -- SQL/command injection, path traversal, auth mistakes,
+# unsafe deserialization. Findings appear in the repo's Security tab. This is
+# the deepest check in the suite and the most valuable for a high-profile
+# target.
+#
+# It runs on every push to main and on a weekly schedule (to catch newly
+# disclosed query patterns against unchanged code). It deliberately does NOT
+# run on pull requests: most PRs here come from forks, whose read-only token
+# cannot publish results, which would produce confusing failures. To scan pull
+# requests too, a maintainer can instead enable CodeQL "default setup" in
+# Settings -> Security -> Code scanning (one toggle, no file needed) -- see
+# docs/security-ci.md.
+
+name: CodeQL
+
+on:
+  push:
+    branches: [main]
+  schedule:
+    # Weekly, Monday 06:00 UTC.
+    - cron: '0 6 * * 1'
+  workflow_dispatch:
+
+permissions: {}
+
+concurrency:
+  group: codeql-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  analyze:
+    name: Analyze (${{ matrix.language }})
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      security-events: write  # publish results to the Security tab
+    strategy:
+      fail-fast: false
+      matrix:
+        # Both are interpreted, so CodeQL needs no build step (build-mode none).
+        language: [python, javascript-typescript]
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Initialize CodeQL
+        uses: github/codeql-action/init@03e4368ac7daa2bd82b3e85262f3bf87ee112f57  # v3.36.0
+        with:
+          languages: ${{ matrix.language }}
+          build-mode: none
+
+      - name: Perform CodeQL analysis
+        uses: github/codeql-action/analyze@03e4368ac7daa2bd82b3e85262f3bf87ee112f57  # v3.36.0
+        with:
+          category: "/language:${{ matrix.language }}"

.github/workflows/container-scan.yml

@@ -0,0 +1,52 @@
+# Container security: Dockerfile lint
+#
+# Purpose: the Docker image is how most people run Odysseus, so it is part of
+# the attack surface. hadolint lints the Dockerfile for mistakes and insecure
+# patterns (running as root longer than needed, unpinned base image, bad apt
+# usage). Blocking.
+#
+# The image vulnerability scan (Trivy, advisory) lives in its own file,
+# container-trivy.yml. Keeping it separate lets that advisory scan be
+# path-filtered and held to a read-only token on pull requests without
+# weakening this blocking gate, which must always report so a required check
+# never hangs.
+#
+# Note: a separate open PR (#120) proposes a local `scripts/scan_image.py`.
+# This job is complementary -- it is a CI gate, not a script a contributor has
+# to remember to run.
+
+name: Container scan
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+permissions: {}
+
+concurrency:
+  group: container-scan-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  hadolint:
+    name: hadolint (Dockerfile lint)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Lint Dockerfile
+        uses: hadolint/hadolint-action@2332a7b74a6de0dda2e2221d575162eba76ba5e5  # v3.3.0
+        with:
+          dockerfile: Dockerfile
+          # DL3008: pinning apt package versions is impractical on a -slim base
+          # image. Debian purges old package versions from its repos, so a
+          # pinned version breaks future rebuilds. The base image itself is
+          # what should be pinned (tracked by Dependabot's docker ecosystem).
+          ignore: DL3008

.github/workflows/container-trivy.yml

@@ -0,0 +1,125 @@
+# Container image vulnerability scan (advisory)
+#
+# Trivy builds the application image and scans it for known-vulnerable OS and
+# Python packages. Advisory only -- it reports findings to the repo's Security
+# tab without blocking a merge, because the image inevitably contains
+# already-known CVEs in upstream packages that are not this project's bug.
+#
+# Split from the Dockerfile lint (container-scan.yml) for two reasons:
+#
+#   - Least privilege. The image build runs Dockerfile instructions, which on a
+#     pull request are attacker-influenceable. That path (the `scan` job) is
+#     held to a read-only token and never publishes results. Only `publish`,
+#     which runs on push to main (curated, fast-forwarded from reviewed dev),
+#     gets security-events:write to upload SARIF.
+#   - Cost. Docs-only changes do not rebuild the image (paths-ignore below),
+#     matching docker-publish.yml. hadolint stays on the broad trigger in
+#     container-scan.yml so the blocking gate always reports.
+
+name: Container scan (Trivy)
+
+on:
+  pull_request:
+    paths-ignore:
+      - '**.md'
+      - 'docs/**'
+      - '.github/ISSUE_TEMPLATE/**'
+  push:
+    branches: [main]
+    paths-ignore:
+      - '**.md'
+      - 'docs/**'
+      - '.github/ISSUE_TEMPLATE/**'
+  workflow_dispatch:
+
+permissions: {}
+
+concurrency:
+  group: container-trivy-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  # Pull requests and manual runs: build and scan under a read-only token.
+  # The build executes PR-supplied Dockerfile instructions, so this job must
+  # not hold any write scope, and it does not upload to the Security tab.
+  scan:
+    name: Trivy (image scan, advisory)
+    if: github.event_name != 'push'
+    runs-on: ubuntu-latest
+    # Advisory: a CVE in an upstream package must not block a PR.
+    continue-on-error: true
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Set up Buildx
+        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5  # v4.1.0
+
+      # Build without pushing so a broken Dockerfile is caught here, and the
+      # exact image we ship is what gets scanned.
+      - name: Build image
+        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf  # v7.2.0
+        with:
+          context: .
+          push: false
+          load: true
+          tags: odysseus:ci
+
+      - name: Scan image with Trivy
+        uses: aquasecurity/trivy-action@ed142fd0673e97e23eac54620cfb913e5ce36c25  # v0.36.0
+        with:
+          image-ref: odysseus:ci
+          format: table
+          ignore-unfixed: true
+        env:
+          # Pin the vuln DB source to GHCR to avoid rate-limited Docker Hub
+          # mirrors that flake on shared runners.
+          TRIVY_DB_REPOSITORY: ghcr.io/aquasecurity/trivy-db:2
+
+  # Push to main only: build, scan, and publish SARIF to the Security tab.
+  # This is the only path that runs trusted code, so it is the only one granted
+  # security-events:write.
+  publish:
+    name: Trivy (image scan + SARIF upload)
+    if: github.event_name == 'push'
+    runs-on: ubuntu-latest
+    continue-on-error: true
+    permissions:
+      contents: read
+      security-events: write  # upload SARIF to the Security tab
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Set up Buildx
+        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5  # v4.1.0
+
+      - name: Build image
+        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf  # v7.2.0
+        with:
+          context: .
+          push: false
+          load: true
+          tags: odysseus:ci
+
+      - name: Scan image with Trivy
+        uses: aquasecurity/trivy-action@ed142fd0673e97e23eac54620cfb913e5ce36c25  # v0.36.0
+        with:
+          image-ref: odysseus:ci
+          format: sarif
+          output: trivy-results.sarif
+          ignore-unfixed: true
+        env:
+          TRIVY_DB_REPOSITORY: ghcr.io/aquasecurity/trivy-db:2
+
+      - name: Upload Trivy results
+        uses: github/codeql-action/upload-sarif@03e4368ac7daa2bd82b3e85262f3bf87ee112f57  # v3.36.0
+        with:
+          sarif_file: trivy-results.sarif
+          category: trivy-image

.github/workflows/dependency-review.yml

@@ -0,0 +1,71 @@
+# Supply-chain review
+#
+# Purpose: defend against "side-chain" / supply-chain attacks -- a pull request
+# that adds (or bumps) a dependency to a version with a known vulnerability or a
+# disallowed license. Two layers:
+#
+#   - dependency-review: runs ONLY on pull requests. It compares the
+#     dependencies before and after the PR and blocks the merge if the change
+#     pulls in a package with a known security advisory. This is the gate.
+#   - pip-audit: scans the project's current Python requirements against the
+#     advisory database. Advisory only (it never blocks a merge), because it can
+#     flag a pre-existing issue in an already-shipped dependency.
+
+name: Dependency review
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+# Default-deny token; jobs grant only read access.
+permissions: {}
+
+concurrency:
+  group: dependency-review-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  dependency-review:
+    name: dependency-review (PR gate)
+    # Only meaningful on a pull request -- it needs a base..head diff to review.
+    if: github.event_name == 'pull_request'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Review dependency changes
+        uses: actions/dependency-review-action@a1d282b36b6f3519aa1f3fc636f609c47dddb294  # v5.0.0
+        with:
+          # Fail the PR on any newly introduced moderate-or-worse advisory.
+          fail-on-severity: moderate
+
+  pip-audit:
+    name: pip-audit (advisory)
+    runs-on: ubuntu-latest
+    # Advisory: report known-vulnerable Python deps without blocking the merge.
+    continue-on-error: true
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6.2.0
+        with:
+          python-version: '3.12'
+
+      - name: Run pip-audit on requirements
+        run: |
+          set -euo pipefail
+          pip install pip-audit==2.10.0
+          pip-audit -r requirements.txt -r requirements-optional.txt --strict

.github/workflows/docker-publish.yml

@@ -0,0 +1,140 @@
+name: ci / docker publish
+
+# Build the Odysseus image and publish to GHCR.
+#   push to main -> :latest, :X.Y.Z            (curated release; main is fast-forwarded at releases)
+#   push to dev  -> :dev,    :X.Y.Z-dev.<sha>  (rolling dev + an immutable, traceable pin)
+# Multi-arch (linux/amd64 + linux/arm64): each arch builds on its own native
+# runner and pushes by digest, then a merge job stitches the digests into one
+# manifest list and applies the tags (faster + cleaner than QEMU emulation).
+# Registry: ghcr.io/<owner>/<repo>.
+
+on:
+  push:
+    branches: [dev, main]
+    paths-ignore:
+      - '**.md'
+      - 'docs/**'
+      - '.github/ISSUE_TEMPLATE/**'
+
+concurrency:
+  group: docker-publish-${{ github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+env:
+  REGISTRY: ghcr.io
+  IMAGE_NAME: ${{ github.repository }}
+
+jobs:
+  build:
+    name: build (${{ matrix.arch }})
+    runs-on: ${{ matrix.runner }}
+    permissions:
+      contents: read
+      packages: write
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - platform: linux/amd64
+            arch: amd64
+            runner: ubuntu-latest
+          - platform: linux/arm64
+            arch: arm64
+            runner: ubuntu-24.04-arm
+    steps:
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6.0.3
+        with:
+          persist-credentials: false
+      - name: Set up Buildx
+        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5  # v4.1.0
+      - name: Log in to GHCR
+        uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee  # v4.2.0
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Build and push by digest
+        id: build
+        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf  # v7.2.0
+        with:
+          context: .
+          platforms: ${{ matrix.platform }}
+          outputs: type=image,name=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true
+          cache-from: type=gha,scope=${{ matrix.arch }}
+          cache-to: type=gha,mode=max,scope=${{ matrix.arch }}
+      - name: Export digest
+        run: |
+          mkdir -p /tmp/digests
+          digest="${{ steps.build.outputs.digest }}"
+          touch "/tmp/digests/${digest#sha256:}"
+      - name: Upload digest
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a  # v7.0.1
+        with:
+          name: digest-${{ matrix.arch }}
+          path: /tmp/digests/*
+          if-no-files-found: error
+          retention-days: 1
+
+  merge:
+    name: merge manifest + tag
+    runs-on: ubuntu-latest
+    needs: build
+    permissions:
+      contents: read
+      packages: write
+    steps:
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6.0.3
+        with:
+          persist-credentials: false
+      - name: Read APP_VERSION + short sha
+        id: ver
+        run: |
+          v=$(grep -E '^APP_VERSION' src/constants.py | head -1 | sed -E 's/.*"([^"]+)".*/\1/')
+          [ -n "$v" ] || { echo "APP_VERSION not found"; exit 1; }
+          echo "version=$v" >> "$GITHUB_OUTPUT"
+          echo "short=${GITHUB_SHA::7}" >> "$GITHUB_OUTPUT"
+      - name: Download digests
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c  # v8.0.1
+        with:
+          path: /tmp/digests
+          pattern: digest-*
+          merge-multiple: true
+      - name: Set up Buildx
+        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5  # v4.1.0
+      - name: Log in to GHCR
+        uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee  # v4.2.0
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Compute tags
+        id: meta
+        uses: docker/metadata-action@80c7e94dd9b9319bd5eb7a0e0fe9291e23a2a2e9  # v6.1.0
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            type=raw,value=latest,enable=${{ github.ref == 'refs/heads/main' }}
+            type=raw,value=${{ steps.ver.outputs.version }},enable=${{ github.ref == 'refs/heads/main' }}
+            type=raw,value=dev,enable=${{ github.ref == 'refs/heads/dev' }}
+            type=raw,value=${{ steps.ver.outputs.version }}-dev.${{ steps.ver.outputs.short }},enable=${{ github.ref == 'refs/heads/dev' }}
+      - name: Create manifest list + push tags
+        working-directory: /tmp/digests
+        run: |
+          tags=$(jq -cr '.tags | map("-t " + .) | join(" ")' <<< "$DOCKER_METADATA_OUTPUT_JSON")
+          digests=$(printf "${REGISTRY}/${IMAGE_NAME}@sha256:%s " *)
+          # word-splitting is intended: $tags and $digests each expand to multiple args
+          # shellcheck disable=SC2086
+          docker buildx imagetools create $tags $digests
+        env:
+          REGISTRY: ${{ env.REGISTRY }}
+          IMAGE_NAME: ${{ env.IMAGE_NAME }}
+      - name: Inspect
+        run: |
+          if [ "$GITHUB_REF" = "refs/heads/main" ]; then ref=latest; else ref=dev; fi
+          docker buildx imagetools inspect "${REGISTRY}/${IMAGE_NAME}:${ref}"
+        env:
+          REGISTRY: ${{ env.REGISTRY }}
+          IMAGE_NAME: ${{ env.IMAGE_NAME }}

.github/workflows/issue-description-check.yml

@@ -14,10 +14,11 @@ jobs:
     # Skip bots (Dependabot, release-drafter, etc.)
     if: ${{ github.event.issue.user.type != 'Bot' }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6.0.3
         with:
           sparse-checkout: .github/scripts
+          persist-credentials: false
 
-      - uses: actions/github-script@v7
+      - uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3  # v9.0.0
         with:
           script: return require('./.github/scripts/check-issue-description.js')({github, context, core})

.github/workflows/pr-description-check.yml

@@ -1,28 +1,109 @@
-name: ci / PR description check
+name: ci / PR checks
 
 on:
-  pull_request_target:
-    types: [opened, edited, synchronize, reopened]
+  # pull_request_target runs in the base-repo context (has secrets) so the check
+  # works on fork PRs. Safe here: the checkout pins to the base branch (no fork
+  # code runs) and the scripts only read context.payload and call the GitHub API.
+  pull_request_target:  # zizmor: ignore[dangerous-triggers]
+    types: [opened, edited, synchronize, reopened, ready_for_review]
 
-# pull_request_target runs in the base-repo context (has secrets).
-# The checkout below pins to the base branch so no fork code is executed.
-# The script only reads context.payload and calls the GitHub API.
-permissions:
-  issues: write
-  pull-requests: write
+# Default-deny at the workflow level; each job opts into only the scopes it needs.
+# Note: modifying a PR's labels/comments needs pull-requests:write even though the
+# REST path is under /issues/{n}/...; issues:write alone returns 403 on PRs.
+permissions: {}
 
 jobs:
   check-description:
     name: Check PR description
     runs-on: ubuntu-latest
-    # Skip bots — they open PRs programmatically and have their own process.
+    permissions:
+      contents: read
+      pull-requests: write
+      issues: write
+    # Skip bots: they open PRs programmatically and have their own process.
     if: github.event.pull_request.user.type != 'Bot'
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6.0.3
         with:
           ref: ${{ github.base_ref }}
           sparse-checkout: .github/scripts
+          persist-credentials: false
 
-      - uses: actions/github-script@v7
+      - uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3  # v9.0.0
         with:
           script: return require('./.github/scripts/check-pr-description.js')({github, context, core})
+
+  check-title:
+    name: Check PR title (Conventional Commits)
+    runs-on: ubuntu-latest
+    permissions: {}
+    # Skip bots: they open PRs programmatically and have their own process.
+    if: github.event.pull_request.user.type != 'Bot'
+    steps:
+      - uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3  # v9.0.0
+        with:
+          script: |
+            const title = context.payload.pull_request.title || "";
+            // Conventional Commits: type(optional-scope)(optional !): summary
+            const re = /^(feat|fix|docs|style|refactor|perf|test|build|ci|chore|revert)(\([\w .\/-]+\))?!?: .+/;
+            if (!re.test(title)) {
+              core.setFailed(
+                `PR title is not in Conventional Commits format:\n  "${title}"\n\n` +
+                `Expected: type(scope): summary\n` +
+                `Example:  fix(search): handle empty query\n` +
+                `Types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert.`
+              );
+            } else {
+              core.info(`PR title OK: ${title}`);
+            }
+
+  check-mergeable:
+    name: Flag unmergeable PRs
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+      issues: write
+    # Skip bots: they open PRs programmatically and have their own process.
+    if: github.event.pull_request.user.type != 'Bot'
+    steps:
+      - uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3  # v9.0.0
+        with:
+          script: |
+            const repo = { owner: context.repo.owner, repo: context.repo.repo };
+            const number = context.payload.pull_request.number;
+            const READY = "ready for review";
+            const CONFLICT = "merge conflict";
+
+            // Ensure the conflict label exists (red). Ignore if already present.
+            try {
+              await github.rest.issues.getLabel({ ...repo, name: CONFLICT });
+            } catch {
+              await github.rest.issues.createLabel({
+                ...repo, name: CONFLICT, color: "B60205",
+                description: "Conflicts with the base branch; needs a rebase before review.",
+              }).catch(() => {});
+            }
+
+            // mergeable is computed asynchronously and is often null right after
+            // an event, so poll a few times until GitHub has resolved it.
+            let pr = null;
+            for (let i = 0; i < 5; i++) {
+              const { data } = await github.rest.pulls.get({ ...repo, pull_number: number });
+              if (data.mergeable !== null) { pr = data; break; }
+              await new Promise(r => setTimeout(r, 3000));
+            }
+            if (!pr || pr.draft) return;
+            const labels = pr.labels.map(l => l.name);
+
+            if (pr.mergeable === false) {
+              if (labels.includes(READY)) {
+                await github.rest.issues.removeLabel({ ...repo, issue_number: number, name: READY }).catch(() => {});
+              }
+              if (!labels.includes(CONFLICT)) {
+                await github.rest.issues.addLabels({ ...repo, issue_number: number, labels: [CONFLICT] });
+              }
+            } else if (pr.mergeable === true) {
+              if (labels.includes(CONFLICT)) {
+                await github.rest.issues.removeLabel({ ...repo, issue_number: number, name: CONFLICT }).catch(() => {});
+              }
+            }

.github/workflows/secret-scan.yml

@@ -0,0 +1,60 @@
+# Secret scanning
+#
+# Purpose: stop credentials (API keys, tokens, passwords, private keys) from
+# ever living in the Git history. Odysseus deliberately keeps real secrets in
+# files that are gitignored (.env, data/), but a slip in a future commit -- or a
+# malicious pull request that sneaks one in -- would otherwise go unnoticed.
+# This job reads the repository and the full commit history and fails if it
+# finds anything that looks like a secret.
+#
+# It runs the official gitleaks BINARY directly (pinned to an exact version and
+# verified against the project's published SHA-256 checksum) rather than the
+# gitleaks GitHub Action, because the Action asks for a paid license on
+# organization-owned repos. The binary is free and behaves identically.
+
+name: Secret scan
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+# Start with zero permissions; the single job opts back in to read-only.
+permissions: {}
+
+concurrency:
+  group: secret-scan-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  gitleaks:
+    name: gitleaks
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          # Full history so a secret committed in an earlier commit (and later
+          # deleted) is still caught -- deletion does not remove it from Git.
+          fetch-depth: 0
+          persist-credentials: false
+
+      # Pinned version + checksum so a tampered release binary cannot run here.
+      # Bump VERSION/SHA256 together; the checksum comes from the matching
+      # gitleaks_<version>_checksums.txt on the GitHub release.
+      - name: Run gitleaks (pinned, checksum-verified)
+        env:
+          GITLEAKS_VERSION: 8.30.1
+          GITLEAKS_SHA256: 551f6fc83ea457d62a0d98237cbad105af8d557003051f41f3e7ca7b3f2470eb
+        run: |
+          set -euo pipefail
+          TARBALL="gitleaks_${GITLEAKS_VERSION}_linux_x64.tar.gz"
+          curl -fsSL -o "${TARBALL}" \
+            "https://github.com/gitleaks/gitleaks/releases/download/v${GITLEAKS_VERSION}/${TARBALL}"
+          echo "${GITLEAKS_SHA256}  ${TARBALL}" | sha256sum -c -
+          tar -xzf "${TARBALL}" gitleaks
+          # Scan the whole history. Findings print to the log and fail the job.
+          ./gitleaks git --no-banner --redact --verbose .

.github/workflows/workflow-security.yml

@@ -0,0 +1,80 @@
+# Workflow security (CI that audits the CI)
+#
+# Purpose: the GitHub Actions workflows themselves are an attack surface. A
+# poorly written workflow can leak the repository token, run attacker-supplied
+# code from a pull request, or pull in a tampered third-party action. These two
+# tools check every workflow file in this repo for those mistakes:
+#
+#   - actionlint: catches workflow syntax errors and shell-script bugs inside
+#     `run:` steps before they reach main.
+#   - zizmor: a security linter for Actions. Flags template-injection holes,
+#     unpinned actions, credential persistence, and over-broad token
+#     permissions -- exactly the patterns the rest of this CI is built to avoid.
+#
+# Add this early: it then audits every workflow added after it.
+
+name: Workflow security
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+# Default-deny token; each job grants only read access to the code.
+permissions: {}
+
+concurrency:
+  group: workflow-security-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  actionlint:
+    name: actionlint
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      # Pinned version + checksum so a tampered binary cannot run here.
+      - name: Run actionlint (pinned, checksum-verified)
+        env:
+          ACTIONLINT_VERSION: 1.7.12
+          ACTIONLINT_SHA256: 8aca8db96f1b94770f1b0d72b6dddcb1ebb8123cb3712530b08cc387b349a3d8
+        run: |
+          set -euo pipefail
+          TARBALL="actionlint_${ACTIONLINT_VERSION}_linux_amd64.tar.gz"
+          curl -fsSL -o "${TARBALL}" \
+            "https://github.com/rhysd/actionlint/releases/download/v${ACTIONLINT_VERSION}/${TARBALL}"
+          echo "${ACTIONLINT_SHA256}  ${TARBALL}" | sha256sum -c -
+          tar -xzf "${TARBALL}" actionlint
+          ./actionlint -color
+
+  zizmor:
+    name: zizmor (Actions SAST)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6.2.0
+        with:
+          python-version: '3.12'
+
+      # Pinned zizmor release. --offline keeps the audit hermetic (no network
+      # calls about the actions it inspects); --min-severity=low surfaces
+      # everything so nothing slips through under the gate.
+      - name: Run zizmor
+        run: |
+          set -euo pipefail
+          pip install zizmor==1.25.2
+          zizmor --offline --min-severity=low .github/workflows/

.gitignore

@@ -89,3 +89,4 @@ docs/windows-port/
 compound.config.json
 *.error.log
 _scratch/
+/odysseus/

CONTRIBUTING.md

@@ -94,6 +94,18 @@ Before submitting any change that affects what the app looks like — buttons, i
 
 If you are unsure whether a change is "visual," it is. Default to attaching a screenshot.
 
+## Code conventions
+
+Don't hardcode values that the project already exposes through a constant or a helper. Hardcoded literals drift out of sync, break on non-default deployments, and reintroduce bugs we've already fixed.
+
+- **Filesystem paths:** never build writable paths from `Path(__file__)...` into the source tree, hardcode `/app/...`, or use a relative `"data/..."` string. Every persisted file and directory has a named constant in `src/constants.py` (for example `AUTH_FILE`, `USER_PREFS_FILE`, `SETTINGS_FILE`, `TTS_CACHE_DIR`, `CHROMA_DIR`). Import and use that named constant; do not re-derive the path locally with `os.path.join(DATA_DIR, "x.json")` or `DATA_DIR / "x.json"`. `DATA_DIR` is the single place that reads `ODYSSEUS_DATA_DIR`, so use it directly only for dynamic paths that have no fixed name (for example per-owner files). If a data file or directory has no constant yet, add one to `src/constants.py`. The source tree is read-only in Docker and `/app/...` does not exist on native runs; guard directory creation so an unwritable path degrades gracefully instead of crashing at import.
+- **Internal API / loopback URLs:** don't hardcode `http://localhost:7000`. Use `internal_api_base()` from `src.constants` (it honors `ODYSSEUS_INTERNAL_BASE` / `APP_PORT`).
+- **Ports, limits, model lists, and similar:** reuse the existing constant if one exists; if it doesn't and the value is used in more than one place, add a constant rather than copying the literal.
+
+If you need a value that has no constant or helper yet, add it to `src/constants.py` (the single source of truth for paths and config; `core/constants.py` only re-exports it for backward compatibility) and import it, rather than repeating a literal across files.
+
+**Commits:** use [Conventional Commits](https://www.conventionalcommits.org), `type(scope): summary` (e.g. `fix(search): ...`, `feat(notes): ...`, `docs(contributing): ...`). Common types: `fix`, `feat`, `refactor`, `docs`, `test`, `chore`, `ci`. Keep the subject short and imperative; put the "why" in the body when it isn't obvious.
+
 ## Issue Reports
 
 For bugs, include:

LICENSE

@@ -1,21 +1,235 @@
-MIT License
+GNU AFFERO GENERAL PUBLIC LICENSE
+Version 3, 19 November 2007
 
-Copyright (c) 2025 Odysseus Contributors
+Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
 
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
+Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed.
 
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
+                            Preamble
 
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+The GNU Affero General Public License is a free, copyleft license for software and other kinds of works, specifically designed to ensure cooperation with the community in the case of network server software.
+
+The licenses for most software and other practical works are designed to take away your freedom to share and change the works.  By contrast, our General Public Licenses are intended to guarantee your freedom to share and change all versions of a program--to make sure it remains free software for all its users.
+
+When we speak of free software, we are referring to freedom, not price.  Our General Public Licenses are designed to make sure that you have the freedom to distribute copies of free software (and charge for them if you wish), that you receive source code or can get it if you want it, that you can change the software or use pieces of it in new free programs, and that you know you can do these things.
+
+Developers that use our General Public Licenses protect your rights with two steps: (1) assert copyright on the software, and (2) offer you this License which gives you legal permission to copy, distribute and/or modify the software.
+
+A secondary benefit of defending all users' freedom is that improvements made in alternate versions of the program, if they receive widespread use, become available for other developers to incorporate.  Many developers of free software are heartened and encouraged by the resulting cooperation.  However, in the case of software used on network servers, this result may fail to come about. The GNU General Public License permits making a modified version and letting the public access it on a server without ever releasing its source code to the public.
+
+The GNU Affero General Public License is designed specifically to ensure that, in such cases, the modified source code becomes available to the community.  It requires the operator of a network server to provide the source code of the modified version running there to the users of that server.  Therefore, public use of a modified version, on a publicly accessible server, gives the public access to the source code of the modified version.
+
+An older license, called the Affero General Public License and published by Affero, was designed to accomplish similar goals.  This is a different license, not a version of the Affero GPL, but Affero has released a new version of the Affero GPL which permits relicensing under this license.
+
+The precise terms and conditions for copying, distribution and modification follow.
+
+                       TERMS AND CONDITIONS
+
+0. Definitions.
+
+"This License" refers to version 3 of the GNU Affero General Public License.
+
+"Copyright" also means copyright-like laws that apply to other kinds of works, such as semiconductor masks.
+
+"The Program" refers to any copyrightable work licensed under this License.  Each licensee is addressed as "you".  "Licensees" and "recipients" may be individuals or organizations.
+
+To "modify" a work means to copy from or adapt all or part of the work in a fashion requiring copyright permission, other than the making of an exact copy.  The resulting work is called a "modified version" of the earlier work or a work "based on" the earlier work.
+
+A "covered work" means either the unmodified Program or a work based on the Program.
+
+To "propagate" a work means to do anything with it that, without permission, would make you directly or secondarily liable for infringement under applicable copyright law, except executing it on a computer or modifying a private copy.  Propagation includes copying, distribution (with or without modification), making available to the public, and in some countries other activities as well.
+
+To "convey" a work means any kind of propagation that enables other parties to make or receive copies.  Mere interaction with a user through a computer network, with no transfer of a copy, is not conveying.
+
+An interactive user interface displays "Appropriate Legal Notices" to the extent that it includes a convenient and prominently visible feature that (1) displays an appropriate copyright notice, and (2) tells the user that there is no warranty for the work (except to the extent that warranties are provided), that licensees may convey the work under this License, and how to view a copy of this License.  If the interface presents a list of user commands or options, such as a menu, a prominent item in the list meets this criterion.
+
+1. Source Code.
+The "source code" for a work means the preferred form of the work for making modifications to it.  "Object code" means any non-source form of a work.
+
+A "Standard Interface" means an interface that either is an official standard defined by a recognized standards body, or, in the case of interfaces specified for a particular programming language, one that is widely used among developers working in that language.
+
+The "System Libraries" of an executable work include anything, other than the work as a whole, that (a) is included in the normal form of packaging a Major Component, but which is not part of that Major Component, and (b) serves only to enable use of the work with that Major Component, or to implement a Standard Interface for which an implementation is available to the public in source code form.  A "Major Component", in this context, means a major essential component (kernel, window system, and so on) of the specific operating system (if any) on which the executable work runs, or a compiler used to produce the work, or an object code interpreter used to run it.
+
+The "Corresponding Source" for a work in object code form means all the source code needed to generate, install, and (for an executable work) run the object code and to modify the work, including scripts to control those activities.  However, it does not include the work's System Libraries, or general-purpose tools or generally available free programs which are used unmodified in performing those activities but which are not part of the work.  For example, Corresponding Source includes interface definition files associated with source files for the work, and the source code for shared libraries and dynamically linked subprograms that the work is specifically designed to require, such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+The Corresponding Source need not include anything that users can regenerate automatically from other parts of the Corresponding Source.
+
+The Corresponding Source for a work in source code form is that same work.
+
+2. Basic Permissions.
+All rights granted under this License are granted for the term of copyright on the Program, and are irrevocable provided the stated conditions are met.  This License explicitly affirms your unlimited permission to run the unmodified Program.  The output from running a covered work is covered by this License only if the output, given its content, constitutes a covered work.  This License acknowledges your rights of fair use or other equivalent, as provided by copyright law.
+
+You may make, run and propagate covered works that you do not convey, without conditions so long as your license otherwise remains in force.  You may convey covered works to others for the sole purpose of having them make modifications exclusively for you, or provide you with facilities for running those works, provided that you comply with the terms of this License in conveying all material for which you do not control copyright.  Those thus making or running the covered works for you must do so exclusively on your behalf, under your direction and control, on terms that prohibit them from making any copies of your copyrighted material outside their relationship with you.
+
+Conveying under any other circumstances is permitted solely under the conditions stated below.  Sublicensing is not allowed; section 10 makes it unnecessary.
+
+3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+No covered work shall be deemed part of an effective technological measure under any applicable law fulfilling obligations under article 11 of the WIPO copyright treaty adopted on 20 December 1996, or similar laws prohibiting or restricting circumvention of such measures.
+
+When you convey a covered work, you waive any legal power to forbid circumvention of technological measures to the extent such circumvention is effected by exercising rights under this License with respect to the covered work, and you disclaim any intention to limit operation or modification of the work as a means of enforcing, against the work's users, your or third parties' legal rights to forbid circumvention of technological measures.
+
+4. Conveying Verbatim Copies.
+You may convey verbatim copies of the Program's source code as you receive it, in any medium, provided that you conspicuously and appropriately publish on each copy an appropriate copyright notice; keep intact all notices stating that this License and any non-permissive terms added in accord with section 7 apply to the code; keep intact all notices of the absence of any warranty; and give all recipients a copy of this License along with the Program.
+
+You may charge any price or no price for each copy that you convey, and you may offer support or warranty protection for a fee.
+
+5. Conveying Modified Source Versions.
+You may convey a work based on the Program, or the modifications to produce it from the Program, in the form of source code under the terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is released under this License and any conditions added under section 7.  This requirement modifies the requirement in section 4 to "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this License to anyone who comes into possession of a copy.  This License will therefore apply, along with any applicable section 7 additional terms, to the whole of the work, and all its parts, regardless of how they are packaged.  This License gives no permission to license the work in any other way, but it does not invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display Appropriate Legal Notices; however, if the Program has interactive interfaces that do not display Appropriate Legal Notices, your work need not make them do so.
+
+A compilation of a covered work with other separate and independent works, which are not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program, in or on a volume of a storage or distribution medium, is called an "aggregate" if the compilation and its resulting copyright are not used to limit the access or legal rights of the compilation's users beyond what the individual works permit.  Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate.
+
+6. Conveying Non-Source Forms.
+You may convey a covered work in object code form under the terms of sections 4 and 5, provided that you also convey the machine-readable Corresponding Source under the terms of this License, in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product (including a physical distribution medium), accompanied by the Corresponding Source fixed on a durable physical medium customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product (including a physical distribution medium), accompanied by a written offer, valid for at least three years and valid for as long as you offer spare parts or customer support for that product model, to give anyone who possesses the object code either (1) a copy of the Corresponding Source for all the software in the product that is covered by this License, on a durable physical medium customarily used for software interchange, for a price no more than your reasonable cost of physically performing this conveying of source, or (2) access to copy the Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the written offer to provide the Corresponding Source.  This alternative is allowed only occasionally and noncommercially, and only if you received the object code with such an offer, in accord with subsection 6b.
+
+    d) Convey the object code by offering access from a designated place (gratis or for a charge), and offer equivalent access to the Corresponding Source in the same way through the same place at no further charge.  You need not require recipients to copy the Corresponding Source along with the object code.  If the place to copy the object code is a network server, the Corresponding Source may be on a different server (operated by you or a third party) that supports equivalent copying facilities, provided you maintain clear directions next to the object code saying where to find the Corresponding Source.  Regardless of what server hosts the Corresponding Source, you remain obligated to ensure that it is available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided you inform other peers where the object code and Corresponding Source of the work are being offered to the general public at no charge under subsection 6d.
+
+A separable portion of the object code, whose source code is excluded from the Corresponding Source as a System Library, need not be included in conveying the object code work.
+
+A "User Product" is either (1) a "consumer product", which means any tangible personal property which is normally used for personal, family, or household purposes, or (2) anything designed or sold for incorporation into a dwelling.  In determining whether a product is a consumer product, doubtful cases shall be resolved in favor of coverage.  For a particular product received by a particular user, "normally used" refers to a typical or common use of that class of product, regardless of the status of the particular user or of the way in which the particular user actually uses, or expects or is expected to use, the product.  A product is a consumer product regardless of whether the product has substantial commercial, industrial or non-consumer uses, unless such uses represent the only significant mode of use of the product.
+
+"Installation Information" for a User Product means any methods, procedures, authorization keys, or other information required to install and execute modified versions of a covered work in that User Product from a modified version of its Corresponding Source.  The information must suffice to ensure that the continued functioning of the modified object code is in no case prevented or interfered with solely because modification has been made.
+
+If you convey an object code work under this section in, or with, or specifically for use in, a User Product, and the conveying occurs as part of a transaction in which the right of possession and use of the User Product is transferred to the recipient in perpetuity or for a fixed term (regardless of how the transaction is characterized), the Corresponding Source conveyed under this section must be accompanied by the Installation Information.  But this requirement does not apply if neither you nor any third party retains the ability to install modified object code on the User Product (for example, the work has been installed in ROM).
+
+The requirement to provide Installation Information does not include a requirement to continue to provide support service, warranty, or updates for a work that has been modified or installed by the recipient, or for the User Product in which it has been modified or installed.  Access to a network may be denied when the modification itself materially and adversely affects the operation of the network or violates the rules and protocols for communication across the network.
+
+Corresponding Source conveyed, and Installation Information provided, in accord with this section must be in a format that is publicly documented (and with an implementation available to the public in source code form), and must require no special password or key for unpacking, reading or copying.
+
+7. Additional Terms.
+"Additional permissions" are terms that supplement the terms of this License by making exceptions from one or more of its conditions. Additional permissions that are applicable to the entire Program shall be treated as though they were included in this License, to the extent that they are valid under applicable law.  If additional permissions apply only to part of the Program, that part may be used separately under those permissions, but the entire Program remains governed by this License without regard to the additional permissions.
+
+When you convey a copy of a covered work, you may at your option remove any additional permissions from that copy, or from any part of it.  (Additional permissions may be written to require their own removal in certain cases when you modify the work.)  You may place additional permissions on material, added by you to a covered work, for which you have or can give appropriate copyright permission.
+
+Notwithstanding any other provision of this License, for material you add to a covered work, you may (if authorized by the copyright holders of that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or author attributions in that material or in the Appropriate Legal Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or requiring that modified versions of such material be marked in reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that material by anyone who conveys the material (or modified versions of it) with contractual assumptions of liability to the recipient, for any liability that these contractual assumptions directly impose on those licensors and authors.
+
+All other non-permissive additional terms are considered "further restrictions" within the meaning of section 10.  If the Program as you received it, or any part of it, contains a notice stating that it is governed by this License along with a term that is a further restriction, you may remove that term.  If a license document contains a further restriction but permits relicensing or conveying under this License, you may add to a covered work material governed by the terms of that license document, provided that the further restriction does not survive such relicensing or conveying.
+
+If you add terms to a covered work in accord with this section, you must place, in the relevant source files, a statement of the additional terms that apply to those files, or a notice indicating where to find the applicable terms.
+
+Additional terms, permissive or non-permissive, may be stated in the form of a separately written license, or stated as exceptions; the above requirements apply either way.
+
+8. Termination.
+
+You may not propagate or modify a covered work except as expressly provided under this License.  Any attempt otherwise to propagate or modify it is void, and will automatically terminate your rights under this License (including any patent licenses granted under the third paragraph of section 11).
+
+However, if you cease all violation of this License, then your license from a particular copyright holder is reinstated (a) provisionally, unless and until the copyright holder explicitly and finally terminates your license, and (b) permanently, if the copyright holder fails to notify you of the violation by some reasonable means prior to 60 days after the cessation.
+
+Moreover, your license from a particular copyright holder is reinstated permanently if the copyright holder notifies you of the violation by some reasonable means, this is the first time you have received notice of violation of this License (for any work) from that copyright holder, and you cure the violation prior to 30 days after your receipt of the notice.
+
+Termination of your rights under this section does not terminate the licenses of parties who have received copies or rights from you under this License.  If your rights have been terminated and not permanently reinstated, you do not qualify to receive new licenses for the same material under section 10.
+
+9. Acceptance Not Required for Having Copies.
+
+You are not required to accept this License in order to receive or run a copy of the Program.  Ancillary propagation of a covered work occurring solely as a consequence of using peer-to-peer transmission to receive a copy likewise does not require acceptance.  However, nothing other than this License grants you permission to propagate or modify any covered work.  These actions infringe copyright if you do not accept this License.  Therefore, by modifying or propagating a covered work, you indicate your acceptance of this License to do so.
+
+10. Automatic Licensing of Downstream Recipients.
+
+Each time you convey a covered work, the recipient automatically receives a license from the original licensors, to run, modify and propagate that work, subject to this License.  You are not responsible for enforcing compliance by third parties with this License.
+
+An "entity transaction" is a transaction transferring control of an organization, or substantially all assets of one, or subdividing an organization, or merging organizations.  If propagation of a covered work results from an entity transaction, each party to that transaction who receives a copy of the work also receives whatever licenses to the work the party's predecessor in interest had or could give under the previous paragraph, plus a right to possession of the Corresponding Source of the work from the predecessor in interest, if the predecessor has it or can get it with reasonable efforts.
+
+You may not impose any further restrictions on the exercise of the rights granted or affirmed under this License.  For example, you may not impose a license fee, royalty, or other charge for exercise of rights granted under this License, and you may not initiate litigation (including a cross-claim or counterclaim in a lawsuit) alleging that any patent claim is infringed by making, using, selling, offering for sale, or importing the Program or any portion of it.
+
+11. Patents.
+
+A "contributor" is a copyright holder who authorizes use under this License of the Program or a work on which the Program is based.  The work thus licensed is called the contributor's "contributor version".
+
+A contributor's "essential patent claims" are all patent claims owned or controlled by the contributor, whether already acquired or hereafter acquired, that would be infringed by some manner, permitted by this License, of making, using, or selling its contributor version, but do not include claims that would be infringed only as a consequence of further modification of the contributor version.  For purposes of this definition, "control" includes the right to grant patent sublicenses in a manner consistent with the requirements of this License.
+
+Each contributor grants you a non-exclusive, worldwide, royalty-free patent license under the contributor's essential patent claims, to make, use, sell, offer for sale, import and otherwise run, modify and propagate the contents of its contributor version.
+
+In the following three paragraphs, a "patent license" is any express agreement or commitment, however denominated, not to enforce a patent (such as an express permission to practice a patent or covenant not to sue for patent infringement).  To "grant" such a patent license to a party means to make such an agreement or commitment not to enforce a patent against the party.
+
+If you convey a covered work, knowingly relying on a patent license, and the Corresponding Source of the work is not available for anyone to copy, free of charge and under the terms of this License, through a publicly available network server or other readily accessible means, then you must either (1) cause the Corresponding Source to be so available, or (2) arrange to deprive yourself of the benefit of the patent license for this particular work, or (3) arrange, in a manner consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have actual knowledge that, but for the patent license, your conveying the covered work in a country, or your recipient's use of the covered work in a country, would infringe one or more identifiable patents in that country that you have reason to believe are valid.
+
+If, pursuant to or in connection with a single transaction or arrangement, you convey, or propagate by procuring conveyance of, a covered work, and grant a patent license to some of the parties receiving the covered work authorizing them to use, propagate, modify or convey a specific copy of the covered work, then the patent license you grant is automatically extended to all recipients of the covered work and works based on it.
+
+A patent license is "discriminatory" if it does not include within the scope of its coverage, prohibits the exercise of, or is conditioned on the non-exercise of one or more of the rights that are specifically granted under this License.  You may not convey a covered work if you are a party to an arrangement with a third party that is in the business of distributing software, under which you make payment to the third party based on the extent of your activity of conveying the work, and under which the third party grants, to any of the parties who would receive the covered work from you, a discriminatory patent license (a) in connection with copies of the covered work conveyed by you (or copies made from those copies), or (b) primarily for and in connection with specific products or compilations that contain the covered work, unless you entered into that arrangement, or that patent license was granted, prior to 28 March 2007.
+
+Nothing in this License shall be construed as excluding or limiting any implied license or other defenses to infringement that may otherwise be available to you under applicable patent law.
+
+12. No Surrender of Others' Freedom.
+
+If conditions are imposed on you (whether by court order, agreement or otherwise) that contradict the conditions of this License, they do not excuse you from the conditions of this License.  If you cannot convey a covered work so as to satisfy simultaneously your obligations under this License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you to collect a royalty for further conveying from those to whom you convey the Program, the only way you could satisfy both those terms and this License would be to refrain entirely from conveying the Program.
+
+13. Remote Network Interaction; Use with the GNU General Public License.
+
+Notwithstanding any other provision of this License, if you modify the Program, your modified version must prominently offer all users interacting with it remotely through a computer network (if your version supports such interaction) an opportunity to receive the Corresponding Source of your version by providing access to the Corresponding Source from a network server at no charge, through some standard or customary means of facilitating copying of software.  This Corresponding Source shall include the Corresponding Source for any work covered by version 3 of the GNU General Public License that is incorporated pursuant to the following paragraph.
+
+Notwithstanding any other provision of this License, you have permission to link or combine any covered work with a work licensed under version 3 of the GNU General Public License into a single combined work, and to convey the resulting work.  The terms of this License will continue to apply to the part which is the covered work, but the work with which it is combined will remain governed by version 3 of the GNU General Public License.
+
+14. Revised Versions of this License.
+
+The Free Software Foundation may publish revised and/or new versions of the GNU Affero General Public License from time to time.  Such new versions will be similar in spirit to the present version, but may differ in detail to address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Program specifies that a certain numbered version of the GNU Affero General Public License "or any later version" applies to it, you have the option of following the terms and conditions either of that numbered version or of any later version published by the Free Software Foundation.  If the Program does not specify a version number of the GNU Affero General Public License, you may choose any version ever published by the Free Software Foundation.
+
+If the Program specifies that a proxy can decide which future versions of the GNU Affero General Public License can be used, that proxy's public statement of acceptance of a version permanently authorizes you to choose that version for the Program.
+
+Later license versions may give you additional or different permissions.  However, no additional obligations are imposed on any author or copyright holder as a result of your choosing to follow a later version.
+
+15. Disclaimer of Warranty.
+
+THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+16. Limitation of Liability.
+
+IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
+
+17. Interpretation of Sections 15 and 16.
+
+If the disclaimer of warranty and limitation of liability provided above cannot be given local legal effect according to their terms, reviewing courts shall apply local law that most closely approximates an absolute waiver of all civil liability in connection with the Program, unless a warranty or assumption of liability accompanies a copy of the Program in return for a fee.
+
+END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+If you develop a new program, and you want it to be of the greatest possible use to the public, the best way to achieve this is to make it free software which everyone can redistribute and change under these terms.
+
+To do so, attach the following notices to the program.  It is safest to attach them to the start of each source file to most effectively state the exclusion of warranty; and each file should have at least the "copyright" line and a pointer to where the full notice is found.
+
+     <one line to give the program's name and a brief idea of what it does.>
+     Copyright (C) <year>  <name of author>
+
+     This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
+
+     This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Affero General Public License for more details.
+
+     You should have received a copy of the GNU Affero General Public License along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+If your software can interact with users remotely through a computer network, you should also make sure that it provides a way for users to get its source.  For example, if your program is a web application, its interface could display a "Source" link that leads users to an archive of the code.  There are many ways you could offer source, and different solutions will be better for different programs; see section 13 for the specific requirements.
+
+You should also get your employer (if you work as a programmer) or school, if any, to sign a "copyright disclaimer" for the program, if necessary. For more information on this, and how to apply and follow the GNU AGPL, see <http://www.gnu.org/licenses/>.

README.md

@@ -1,5 +1,7 @@
 # Odysseus
 
+> **Branch note:** `dev` is the default branch and contains the latest development changes, but it may be unstable. For the more stable curated branch, use [`main`](https://github.com/pewdiepie-archdaemon/odysseus/tree/main).
+
 ```
 ───────────────────────────────────────────────
  ⊹ ࣪ ˖ ૮( ˶ᵔ ᵕ ᵔ˶ )っ  Odysseus vers. 1.0
@@ -216,7 +218,7 @@ docker compose exec odysseus sh -lc 'test -e /dev/kfd && test -d /dev/dri && ls
 > the CUDA Toolkit at runtime. If Cookbook logs show `Unable to find cudart
 > library`, `Could NOT find CUDAToolkit`, `CUDA Toolkit not found`, or
 > tensors/layers assigned to CPU, that is a Cookbook/llama.cpp build issue —
-> not a Docker passthrough failure. Re-install the serve engine via
+> not a Docker passthrough failure. Reinstall the serve engine via
 > **Cookbook → Dependencies** to get a CUDA-enabled build.
 >
 > The same split applies to AMD/ROCm: seeing `/dev/kfd` and `/dev/dri` inside
@@ -327,10 +329,16 @@ To expose Odysseus on a local network or Tailscale with HTTPS:
 | Package | Feature unlocked |
 |---------|-----------------|
 | `faster-whisper` | Local speech-to-text (microphone -> text) via the "local" STT provider. |
-| `duckduckgo-search` | DuckDuckGo as a search provider option. |
+| `ddgs` | DuckDuckGo as a search provider option. |
 | `PyMuPDF` | PDF page rendering in the side viewer panel and form-filling. (Note: AGPL-3.0) |
 | `markitdown` | Office/EPUB document text extraction (converts .docx/.xlsx/.pptx/.xls/.epub to Markdown). |
 
+### Outlook / Office 365 email
+Odysseus email accounts currently use IMAP/SMTP username-password auth. Outlook
+and Microsoft 365 generally require OAuth instead, so normal Microsoft mailbox
+passwords will fail. See [docs/email-outlook.md](docs/email-outlook.md) for the
+current limitation and the planned integration direction.
+
 ## Security Notes
 Odysseus is a self-hosted workspace with powerful local tools: shell access, file uploads, model downloads, web research, email/calendar integrations, and API tokens. Treat it like an admin console.
 
@@ -394,6 +402,16 @@ Key settings:
 | `CHROMADB_HOST` | `localhost` | ChromaDB host for vector memory. Docker overrides this to `chromadb`. |
 | `CHROMADB_PORT` | `8100` | ChromaDB port for manual host runs. Docker overrides this to `8000`. |
 | `EMBEDDING_URL` | -- | OpenAI-compatible embeddings endpoint |
+| `ODYSSEUS_CHAT_UPLOAD_MAX_BYTES` | `10485760` | Chat/agent attachment cap in bytes. Raise for larger local PDFs or text documents. |
+| `ODYSSEUS_GALLERY_UPLOAD_MAX_BYTES` | `104857600` | Gallery image upload cap in bytes (100 MB). |
+| `ODYSSEUS_GALLERY_TRANSFORM_UPLOAD_MAX_BYTES` | `26214400` | Gallery transform input cap in bytes (25 MB). |
+| `ODYSSEUS_MEMORY_IMPORT_MAX_BYTES` | `10485760` | Memory import file cap in bytes (10 MB). |
+| `ODYSSEUS_PERSONAL_UPLOAD_MAX_BYTES` | `26214400` | Personal document upload cap in bytes (25 MB). |
+| `ODYSSEUS_EMAIL_COMPOSE_UPLOAD_MAX_BYTES` | `26214400` | Email compose attachment cap in bytes (25 MB). |
+| `ODYSSEUS_STT_MAX_AUDIO_BYTES` | `26214400` | Speech-to-text audio cap in bytes (25 MB). |
+| `ODYSSEUS_ICS_MAX_BYTES` | `10485760` | Calendar `.ics` import cap in bytes (10 MB). |
+
+All upload-limit vars are validated (must be a positive integer) and optional; an invalid value fails fast at startup.
 
 ### Built-in MCP servers (optional setup)
 
@@ -433,7 +451,7 @@ All user data lives in `data/` (gitignored): `app.db` (sessions, messages, docum
 </a>
 
 ## License
-MIT -- see [LICENSE](LICENSE) and [ACKNOWLEDGMENTS.md](ACKNOWLEDGMENTS.md).
+AGPL-3.0-or-later -- see [LICENSE](LICENSE) and [ACKNOWLEDGMENTS.md](ACKNOWLEDGMENTS.md).
 
 ```
                                   |

app.py

@@ -47,15 +47,16 @@ from fastapi.responses import JSONResponse, FileResponse, HTMLResponse
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles
 from starlette.middleware.base import BaseHTTPMiddleware
+from starlette.middleware.gzip import GZipMiddleware
 
 # Core imports
 from core.constants import (
     BASE_DIR, STATIC_DIR, SESSIONS_FILE,
-    REQUEST_TIMEOUT, OPENAI_API_KEY,
+    REQUEST_TIMEOUT, OPENAI_API_KEY, AUTH_FILE,
 )
 from core.database import SessionLocal, ApiToken
-from core.middleware import SecurityHeadersMiddleware
-from core.auth import AuthManager
+from core.middleware import SecurityHeadersMiddleware, is_cors_preflight
+from core.auth import AuthManager, normalize_known_username
 from core.exceptions import (
     SessionNotFoundError, InvalidFileUploadError,
     LLMServiceError, WebSearchError,
@@ -64,6 +65,7 @@ from core.exceptions import (
 import bcrypt as _bcrypt
 
 from src.app_helpers import abs_join
+from src.generated_images import GENERATED_IMAGE_HEADERS, resolve_generated_image_path
 from starlette.responses import RedirectResponse
 
 # ========= LOGGING =========
@@ -103,6 +105,16 @@ app.add_middleware(
     ],
 )
 
+# ========= RESPONSE COMPRESSION (gzip) =========
+# The frontend's text assets (style.css, index.html, the JS bundles) shipped
+# uncompressed on every cold load. gzip cuts CSS/JS/HTML by ~75-85% on the wire
+# with no behavioural change. Starlette's GZipMiddleware excludes
+# `text/event-stream` by default, so the SSE streams (chat, shell, research,
+# model-probe — all served with media_type="text/event-stream") are never
+# compressed or buffered; only complete bodies over minimum_size are. The
+# security-header middleware composes cleanly on top.
+app.add_middleware(GZipMiddleware, minimum_size=1024, compresslevel=6)
+
 # ========= SECURITY HEADERS MIDDLEWARE =========
 app.add_middleware(SecurityHeadersMiddleware)
 
@@ -216,8 +228,16 @@ if AUTH_ENABLED:
         try:
             rows = db.query(ApiToken).filter(ApiToken.is_active == True).all()
             for r in rows:
+                owner_key = normalize_known_username(auth_manager.users, getattr(r, "owner", None))
+                if not owner_key:
+                    logger.warning(
+                        "Ignoring active API token '%s' for unknown auth user '%s'",
+                        getattr(r, "id", ""),
+                        getattr(r, "owner", None),
+                    )
+                    continue
                 scopes = [s.strip() for s in (getattr(r, "scopes", "") or "chat").split(",") if s.strip()]
-                new_map[r.token_prefix].append((r.id, r.token_hash, getattr(r, "owner", None), scopes))
+                new_map[r.token_prefix].append((r.id, r.token_hash, owner_key, scopes))
         finally:
             db.close()
         _token_cache.clear()
@@ -252,6 +272,15 @@ if AUTH_ENABLED:
     class AuthMiddleware(BaseHTTPMiddleware):
         async def dispatch(self, request: Request, call_next):
             path = request.url.path
+            # A genuine CORS preflight (OPTIONS + Access-Control-Request-Method)
+            # carries no credentials by design and must reach CORSMiddleware to be
+            # answered. AuthMiddleware is the outermost middleware, so gating the
+            # preflight on auth 401s it before CORS can respond -- which blocks
+            # every cross-origin browser/WebView client before the real request
+            # is sent. Let real preflights through (only OPTIONS w/ the ACRM
+            # header; never a credentialed request).
+            if is_cors_preflight(request.method, request.headers):
+                return await call_next(request)
             if _is_auth_exempt(path):
                 return await call_next(request)
             # In-process internal-tool token bypass. Used by the agent
@@ -387,13 +416,7 @@ app.mount("/static", _RevalidatingStatic(directory="static"), name="static")
 @app.get("/api/generated-image/{filename}")
 async def serve_generated_image(filename: str, request: Request):
     """Serve generated images from the data directory."""
-    from pathlib import Path
-    import re
-    if not re.match(r'^[a-f0-9]{8,64}\.(png|jpg|jpeg|webp|gif|mp4|mov|webm|mkv|m4v)$', filename):
-        raise HTTPException(status_code=400, detail="Invalid filename")
-    img_path = Path("data/generated_images") / filename
-    if not img_path.exists():
-        raise HTTPException(status_code=404, detail="Image not found")
+    img_path = resolve_generated_image_path(filename)
     # SECURITY: filename is the only key, so anyone who knows / guesses a
     # 12-hex content hash could pull another user's image bytes. Require
     # auth and verify ownership via the gallery row (when one exists).
@@ -429,7 +452,7 @@ async def serve_generated_image(filename: str, request: Request):
     return FileResponse(
         str(img_path),
         media_type=mime,
-        headers={"Cache-Control": "public, max-age=31536000, immutable"},
+        headers=GENERATED_IMAGE_HEADERS,
     )
 
 # ========= YOUTUBE INIT =========
@@ -468,14 +491,20 @@ components = initialize_managers(BASE_DIR, rag_manager)
 session_manager   = components["session_manager"]
 from src.assistant_log import set_session_manager as _set_asst_sm
 _set_asst_sm(session_manager)
+# Set the global session manager singleton (used by core.models.Session.add_message)
+from core.models import set_session_manager_instance
+set_session_manager_instance(session_manager)
+app.state.session_manager = session_manager
 memory_manager    = components["memory_manager"]
 memory_vector     = components.get("memory_vector")
 upload_handler    = components["upload_handler"]
+app.state.upload_handler = upload_handler
 personal_docs_mgr = components["personal_docs_manager"]
 api_key_manager   = components["api_key_manager"]
 preset_manager    = components["preset_manager"]
 chat_processor    = components["chat_processor"]
 research_handler  = components["research_handler"]
+app.state.research_handler = research_handler
 chat_handler      = components["chat_handler"]
 model_discovery   = components["model_discovery"]
 skills_manager    = components["skills_manager"]
@@ -525,9 +554,6 @@ upload_cleanup_task = None
 from routes.emoji_routes import setup_emoji_routes
 app.include_router(setup_emoji_routes())
 
-from routes.workspace_routes import setup_workspace_routes
-app.include_router(setup_workspace_routes())
-
 # Sessions
 from routes.session_routes import setup_session_routes
 session_config = {"REQUEST_TIMEOUT": REQUEST_TIMEOUT, "OPENAI_API_KEY": OPENAI_API_KEY, "SESSIONS_FILE": SESSIONS_FILE}
@@ -572,7 +598,7 @@ app.include_router(setup_preset_routes(preset_manager))
 
 # Diagnostics
 from routes.diagnostics_routes import setup_diagnostics_routes
-app.include_router(setup_diagnostics_routes(rag_manager, rag_available, research_handler))
+app.include_router(setup_diagnostics_routes(rag_manager, rag_available, research_handler, memory_vector))
 
 # Cleanup
 from routes.cleanup_routes import setup_cleanup_routes
@@ -594,6 +620,10 @@ app.include_router(setup_model_routes(model_discovery))
 from routes.copilot_routes import setup_copilot_routes
 app.include_router(setup_copilot_routes())
 
+# ChatGPT Subscription device-flow login
+from routes.chatgpt_subscription_routes import setup_chatgpt_subscription_routes
+app.include_router(setup_chatgpt_subscription_routes())
+
 # TTS
 from routes.tts_routes import setup_tts_routes
 app.include_router(setup_tts_routes(tts_service))
@@ -646,6 +676,9 @@ app.include_router(setup_shell_routes())
 from routes.cookbook_routes import setup_cookbook_routes
 app.include_router(setup_cookbook_routes())
 
+from routes.workspace_routes import setup_workspace_routes
+app.include_router(setup_workspace_routes())
+
 # Hardware model fitting (cookbook "What Fits?" tab)
 from routes.hwfit_routes import setup_hwfit_routes
 app.include_router(setup_hwfit_routes())
@@ -789,6 +822,8 @@ async def serve_backgrounds(request: Request):
 
 @app.get("/login")
 async def serve_login(request: Request):
+    if not AUTH_ENABLED:
+        return RedirectResponse(url="/", status_code=302)
     return _serve_html_with_nonce(request, abs_join(BASE_DIR, "static/login.html"))
 
 @app.get("/api/version")
@@ -916,16 +951,21 @@ async def _startup_event():
     async def _warmup_endpoints():
         try:
             import httpx
-            endpoints = model_discovery.get_endpoints() if model_discovery else []
-            for ep in endpoints[:5]:
-                url = ep.get("url", "").replace("/chat/completions", "/models")
-                if url:
-                    try:
-                        async with httpx.AsyncClient(timeout=5.0) as client:
-                            await client.get(url)
-                        logger.info(f"Warmup ping OK: {url}")
-                    except Exception as e:
-                        logger.debug(f"Warmup ping failed for endpoint: {e}")
+            # model_discovery has no get_endpoints(); that call raised
+            # AttributeError every run and silently disabled warmup/keepalive.
+            # Resolve the /models probe URLs via the real discovery API, off the
+            # event loop since discovery does a blocking port scan.
+            urls = (
+                await asyncio.to_thread(model_discovery.warmup_ping_urls)
+                if model_discovery else []
+            )
+            for url in urls:
+                try:
+                    async with httpx.AsyncClient(timeout=5.0) as client:
+                        await client.get(url)
+                    logger.info(f"Warmup ping OK: {url}")
+                except Exception as e:
+                    logger.debug(f"Warmup ping failed for endpoint: {e}")
         except Exception as e:
             logger.debug(f"Warmup ping skipped: {e}")
 
@@ -948,7 +988,7 @@ async def _startup_event():
         owners = set()
         try:
             import json as _json
-            auth_path = "data/auth.json"
+            auth_path = AUTH_FILE
             with open(auth_path, encoding="utf-8") as f:
                 users = _json.load(f).get("users", {})
             owners.update(users.keys())
@@ -995,7 +1035,7 @@ async def _startup_event():
     # does not make an existing library look empty after auth/account changes.
     try:
         import json as _json
-        auth_path = "data/auth.json"
+        auth_path = AUTH_FILE
         with open(auth_path, encoding="utf-8") as f:
             users = _json.load(f).get("users", {})
         primary_owner = None
@@ -1067,6 +1107,16 @@ async def _startup_event():
                 logger.warning(f"Nightly skill audit failed: {e}")
 
     _startup_tasks.append(asyncio.create_task(_skill_audit_nightly_loop()))
+
+    # Cookbook serve lifecycle — kills scheduler-launched serves whose
+    # window-end has passed. Paired with the cookbook_serve builtin
+    # action; both are no-ops unless a scheduled task actually launches
+    # something with end_after_min set. Removing this line + the
+    # cookbook_serve entry in BUILTIN_ACTIONS + src/cookbook_serve_lifecycle.py
+    # removes the feature.
+    from src.cookbook_serve_lifecycle import cookbook_serve_lifecycle_loop
+    _startup_tasks.append(asyncio.create_task(cookbook_serve_lifecycle_loop()))
+
     logger.info("Application startup complete")
 
 async def _shutdown_event():

companion/pairing.py

@@ -14,6 +14,8 @@ import uuid
 
 import bcrypt
 
+from src.constants import AUTH_FILE
+
 PAIRING_VERSION = 1
 COMPANION_SCOPE = "chat"
 
@@ -61,7 +63,7 @@ def lan_ip_candidates() -> list[str]:
 def find_admin_user() -> str | None:
     """Resolve an admin username from data/auth.json (schema uses is_admin),
     falling back to the first user."""
-    auth_path = os.path.join("data", "auth.json")
+    auth_path = AUTH_FILE
     try:
         with open(auth_path, "r", encoding="utf-8") as f:
             data = json.load(f)

core/auth.py

@@ -30,14 +30,24 @@ DEFAULT_PRIVILEGES = {
     "can_manage_memory": True,
     "max_messages_per_day": 0,
     "allowed_models": [],
+    "allowed_models_restricted": False,
+    # Explicit "block every model" sentinel. An empty `allowed_models` list is
+    # ambiguous — it's also what gets sent when the admin clicks "[All]" — so
+    # we need a dedicated flag to express "this user may use no models at all"
+    # distinctly from "this user has no restriction".
+    "block_all_models": False,
 }
 
 # Admins get everything
 ADMIN_PRIVILEGES = {k: (True if isinstance(v, bool) else (0 if isinstance(v, int) else [])) for k, v in DEFAULT_PRIVILEGES.items()}
+ADMIN_PRIVILEGES["allowed_models_restricted"] = False
+# Admins must never be blocked from using models — the generic dict
+# comprehension above flips every boolean default to True, which would be
+# backwards for this sentinel.
+ADMIN_PRIVILEGES["block_all_models"] = False
 
-DEFAULT_AUTH_PATH = os.path.join(
-    Path(__file__).parent.parent, "data", "auth.json"
-)
+from src.constants import AUTH_FILE
+DEFAULT_AUTH_PATH = AUTH_FILE
 TOKEN_TTL = 60 * 60 * 24 * 7  # 7 days
 
 # Usernames the auth + middleware layer reserve as internal "synthetic owner"
@@ -57,6 +67,14 @@ TOKEN_TTL = 60 * 60 * 24 * 7  # 7 days
 RESERVED_USERNAMES = frozenset({"internal-tool", "api", "demo", "system"})
 
 
+def normalize_known_username(users: Dict[str, Any], username: str | None) -> Optional[str]:
+    """Return a normalized username only when it exists in the auth user map."""
+    key = str(username or "").strip().lower()
+    if not key or key not in users:
+        return None
+    return key
+
+
 def _hash_password(password: str) -> str:
     return bcrypt.hashpw(password.encode("utf-8"), bcrypt.gensalt()).decode("utf-8")
 
@@ -76,12 +94,17 @@ class AuthManager:
         # Guards mutations of self._sessions and the on-disk sessions.json.
         # Validate/create/revoke run concurrently from the FastAPI threadpool.
         self._sessions_lock = threading.RLock()
+        # Guards all mutations of self._config and the on-disk auth.json so
+        # concurrent create/delete/rename/privilege operations don't interleave
+        # and corrupt the user database.
+        self._config_lock = threading.Lock()
         # Guards the first-run setup check-and-write so concurrent requests
         # cannot both observe is_configured==False and both create admin accounts.
         self._setup_lock = threading.Lock()
         self._load()
         self._load_sessions()
         self._migrate_single_user()
+        self._drop_reserved_loaded_users()
         self._migrate_legacy_admin_role()
 
     def _load(self):
@@ -134,7 +157,13 @@ class AuthManager:
     def _migrate_single_user(self):
         """Migrate old single-user format to multi-user format."""
         if "password_hash" in self._config and "users" not in self._config:
-            old_user = self._config.get("username", "admin")
+            old_user = str(self._config.get("username", "admin") or "admin").strip().lower()
+            if old_user in RESERVED_USERNAMES:
+                logger.warning(
+                    "Migrating legacy single-user reserved username '%s' to 'admin'",
+                    old_user,
+                )
+                old_user = "admin"
             old_hash = self._config["password_hash"]
             self._config = {
                 "users": {
@@ -148,6 +177,30 @@ class AuthManager:
             self._save()
             logger.info(f"Migrated single-user auth to multi-user (admin: {old_user})")
 
+    def _drop_reserved_loaded_users(self):
+        """Fail closed for legacy/manual auth rows that collide with sentinels."""
+        users = self._config.get("users")
+        if not isinstance(users, dict):
+            return
+        normalized = {}
+        removed = []
+        for username, data in users.items():
+            key = str(username or "").strip().lower()
+            if not key:
+                continue
+            if key in RESERVED_USERNAMES:
+                removed.append(key)
+                continue
+            normalized[key] = data
+        if removed or normalized != users:
+            self._config["users"] = normalized
+            self._save()
+        if removed:
+            logger.warning(
+                "Removed reserved username(s) from auth config: %s",
+                ", ".join(sorted(set(removed))),
+            )
+
     def _migrate_legacy_admin_role(self):
         """Normalize setup.py's old role='admin' marker to is_admin=True."""
         changed = False
@@ -172,8 +225,9 @@ class AuthManager:
 
     @signup_enabled.setter
     def signup_enabled(self, value: bool):
-        self._config["signup_enabled"] = value
-        self._save()
+        with self._config_lock:
+            self._config["signup_enabled"] = value
+            self._save()
 
     @property
     def is_configured(self) -> bool:
@@ -198,17 +252,18 @@ class AuthManager:
         if username in RESERVED_USERNAMES:
             logger.warning("Refused to create reserved username '%s'", username)
             return False
-        if username in self.users:
-            return False
-        if "users" not in self._config:
-            self._config["users"] = {}
-        self._config["users"][username] = {
-            "password_hash": _hash_password(password),
-            "created": time.time(),
-            "is_admin": is_admin,
-            "privileges": dict(ADMIN_PRIVILEGES if is_admin else DEFAULT_PRIVILEGES),
-        }
-        self._save()
+        with self._config_lock:
+            if username in self.users:
+                return False
+            if "users" not in self._config:
+                self._config["users"] = {}
+            self._config["users"][username] = {
+                "password_hash": _hash_password(password),
+                "created": time.time(),
+                "is_admin": is_admin,
+                "privileges": dict(ADMIN_PRIVILEGES if is_admin else DEFAULT_PRIVILEGES),
+            }
+            self._save()
         logger.info(f"Created user '{username}' (admin={is_admin})")
         return True
 
@@ -221,14 +276,31 @@ class AuthManager:
         their cookie expired naturally (default ~30 days).
         """
         username = username.strip().lower()
-        if username not in self.users:
-            return False
-        if username == requesting_user:
-            return False
-        if not self.users.get(requesting_user, {}).get("is_admin"):
-            return False
-        del self._config["users"][username]
-        self._save()
+        with self._config_lock:
+            if username not in self.users:
+                return False
+            if username == requesting_user:
+                return False
+            if not self.users.get(requesting_user, {}).get("is_admin"):
+                return False
+            # Revoke API bearer tokens before removing the auth row. The bearer
+            # path authenticates from ApiToken rows and does not require the
+            # owner to still exist, so a successful delete must not leave active
+            # rows behind. If the token store is unavailable, fail closed and
+            # keep the user/session state intact so the admin can retry.
+            try:
+                from core.database import get_db_session, ApiToken
+                with get_db_session() as db:
+                    removed_tokens = db.query(ApiToken).filter(ApiToken.owner == username).delete()
+                if removed_tokens:
+                    logger.info(
+                        f"Revoked {removed_tokens} API token(s) owned by deleted user '{username}'"
+                    )
+            except Exception:
+                logger.warning(f"Failed to revoke API tokens for deleted user '{username}'")
+                return False
+            del self._config["users"][username]
+            self._save()
         # Purge all sessions belonging to this user. validate_token doesn't
         # cross-check `self.users`, so without this step a deleted user's
         # cookie keeps authenticating.
@@ -241,18 +313,6 @@ class AuthManager:
                 revoked += 1
         if revoked:
             self._save_sessions()
-        # Also revoke API bearer tokens owned by this user. The bearer auth
-        # path authenticates straight against ApiToken rows and never
-        # re-checks that the owner still exists, so leaving the rows behind
-        # would let a deleted user keep full API access indefinitely.
-        try:
-            from core.database import get_db_session, ApiToken
-            with get_db_session() as db:
-                removed = db.query(ApiToken).filter(ApiToken.owner == username).delete()
-            if removed:
-                logger.info(f"Revoked {removed} API token(s) owned by deleted user '{username}'")
-        except Exception:
-            logger.warning(f"Failed to revoke API tokens for deleted user '{username}'")
         logger.info(f"Deleted user '{username}' (by {requesting_user}); revoked {revoked} active session(s)")
         return True
 
@@ -266,14 +326,15 @@ class AuthManager:
         if new_username in RESERVED_USERNAMES:
             logger.warning("Refused to rename '%s' into reserved username '%s'", old_username, new_username)
             return False
-        if old_username not in self.users:
-            return False
-        if new_username in self.users:
-            return False
-        if not self.users.get(requesting_user, {}).get("is_admin"):
-            return False
-        self._config.setdefault("users", {})[new_username] = self._config["users"].pop(old_username)
-        self._save()
+        with self._config_lock:
+            if old_username not in self.users:
+                return False
+            if new_username in self.users:
+                return False
+            if not self.users.get(requesting_user, {}).get("is_admin"):
+                return False
+            self._config.setdefault("users", {})[new_username] = self._config["users"].pop(old_username)
+            self._save()
 
         renamed_sessions = 0
         with self._sessions_lock:
@@ -311,17 +372,18 @@ class AuthManager:
     def set_privileges(self, username: str, privileges: Dict[str, Any]) -> bool:
         """Update privileges for a user. Can't modify admin privileges."""
         username = username.strip().lower()
-        if username not in self.users:
-            return False
-        if self.users[username].get("is_admin"):
-            return False  # admins always have full access
-        # Only allow known privilege keys
-        current = self.get_privileges(username)
-        for k, v in privileges.items():
-            if k in DEFAULT_PRIVILEGES:
-                current[k] = v
-        self._config["users"][username]["privileges"] = current
-        self._save()
+        with self._config_lock:
+            if username not in self.users:
+                return False
+            if self.users[username].get("is_admin"):
+                return False  # admins always have full access
+            # Only allow known privilege keys
+            current = self.get_privileges(username)
+            for k, v in privileges.items():
+                if k in DEFAULT_PRIVILEGES:
+                    current[k] = v
+            self._config["users"][username]["privileges"] = current
+            self._save()
         logger.info(f"Updated privileges for '{username}': {current}")
         return True
 
@@ -331,8 +393,9 @@ class AuthManager:
             return False
         if not _verify_password(current_password, self.users[username]["password_hash"]):
             return False
-        self._config["users"][username]["password_hash"] = _hash_password(new_password)
-        self._save()
+        with self._config_lock:
+            self._config["users"][username]["password_hash"] = _hash_password(new_password)
+            self._save()
         return True
 
     # ------------------------------------------------------------------
@@ -350,8 +413,9 @@ class AuthManager:
         if username not in self.users:
             return None
         secret = pyotp.random_base32()
-        self._config["users"][username]["totp_secret_pending"] = secret
-        self._save()
+        with self._config_lock:
+            self._config["users"][username]["totp_secret_pending"] = secret
+            self._save()
         return secret
 
     def totp_get_provisioning_uri(self, username: str, secret: str) -> str:
@@ -370,13 +434,14 @@ class AuthManager:
         if not totp.verify(code, valid_window=1):
             return False
         # Enable 2FA
-        self._config["users"][username]["totp_secret"] = secret
-        self._config["users"][username]["totp_enabled"] = True
-        self._config["users"][username].pop("totp_secret_pending", None)
-        # Generate backup codes
-        backup = [secrets.token_hex(4) for _ in range(8)]
-        self._config["users"][username]["totp_backup_codes"] = backup
-        self._save()
+        with self._config_lock:
+            self._config["users"][username]["totp_secret"] = secret
+            self._config["users"][username]["totp_enabled"] = True
+            self._config["users"][username].pop("totp_secret_pending", None)
+            # Generate backup codes
+            backup = [secrets.token_hex(4) for _ in range(8)]
+            self._config["users"][username]["totp_backup_codes"] = backup
+            self._save()
         logger.info(f"2FA enabled for '{username}'")
         return True
 
@@ -395,9 +460,10 @@ class AuthManager:
         # Check backup codes first
         backup = user.get("totp_backup_codes", [])
         if code in backup:
-            backup.remove(code)
-            self._config["users"][username]["totp_backup_codes"] = backup
-            self._save()
+            with self._config_lock:
+                backup.remove(code)
+                self._config["users"][username]["totp_backup_codes"] = backup
+                self._save()
             logger.info(f"Backup code used for '{username}' ({len(backup)} remaining)")
             return True
         totp = pyotp.TOTP(secret)
@@ -408,11 +474,12 @@ class AuthManager:
         username = username.strip().lower()
         if not self.verify_password(username, password):
             return False
-        self._config["users"][username].pop("totp_secret", None)
-        self._config["users"][username].pop("totp_secret_pending", None)
-        self._config["users"][username].pop("totp_backup_codes", None)
-        self._config["users"][username]["totp_enabled"] = False
-        self._save()
+        with self._config_lock:
+            self._config["users"][username].pop("totp_secret", None)
+            self._config["users"][username].pop("totp_secret_pending", None)
+            self._config["users"][username].pop("totp_backup_codes", None)
+            self._config["users"][username]["totp_enabled"] = False
+            self._save()
         logger.info(f"2FA disabled for '{username}'")
         return True
 
@@ -431,6 +498,12 @@ class AuthManager:
         username = username.strip().lower()
         if not self.verify_password(username, password):
             return None
+        return self.create_session_trusted(username)
+
+    def create_session_trusted(self, username: str) -> str:
+        """Issue a session token for an already-verified user.
+        Call only after verify_password (and TOTP if enabled) have passed."""
+        username = username.strip().lower()
         token = secrets.token_hex(32)
         with self._sessions_lock:
             self._sessions[token] = {

core/constants.py

@@ -1,40 +1,12 @@
-# src/constants.py
-"""Application-wide constants and configuration values."""
-import os
+# core/constants.py
+"""Backward-compatible shim — the single source of truth is src/constants.py.
 
-APP_VERSION = "0.9.1"
-
-# Base paths
-BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) + "/"
-STATIC_DIR = os.path.join(BASE_DIR, "static")
-DATA_DIR = os.path.join(BASE_DIR, "data")
-
-# Data file paths
-SESSIONS_FILE = os.path.join(DATA_DIR, "sessions.json")
-MEMORY_FILE = os.path.join(DATA_DIR, "memory.json")
-MEMORY_DOC = os.path.join(DATA_DIR, "memory_doc.md")
-PERSONAL_DIR = os.path.join(DATA_DIR, "personal_docs")
-RUNBOOK_DIR = os.path.join(PERSONAL_DIR, "runbook")
-UPLOAD_DIR = os.path.join(DATA_DIR, "uploads")
-FEATURES_FILE = os.path.join(DATA_DIR, "features.json")
-SETTINGS_FILE = os.path.join(DATA_DIR, "settings.json")
-
-# API Configuration
-MAX_CONTEXT_MESSAGES = 90
-REQUEST_TIMEOUT = 20
-OPENAI_COMPAT_PATH = "/v1/chat/completions"
-
-# Environment variables with defaults
-DEFAULT_HOST = os.getenv("LLM_HOST", "localhost")
-LLM_HOSTS = [h.strip() for h in os.getenv("LLM_HOSTS", "").split(",") if h.strip()]
-OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
-SEARXNG_INSTANCE = os.getenv('SEARXNG_INSTANCE', 'http://localhost:8080')
-
-
-# Cleanup configuration
-CLEANUP_ENABLED = os.getenv("CLEANUP_ENABLED", "True").lower() == "true"
-CLEANUP_INTERVAL_HOURS = int(os.getenv("CLEANUP_INTERVAL_HOURS", "24"))
-
-# Default parameters
-DEFAULT_TEMPERATURE = 1.0
-DEFAULT_MAX_TOKENS = 0
+Historically there were two copies of this module (this one lagged behind at
+APP_VERSION 0.9.1 and was missing the consolidated tool-output constants). To
+kill the drift, this now simply re-exports everything from src.constants so
+there is exactly one place that defines paths and reads ODYSSEUS_DATA_DIR.
+internal_api_base() also lives in src.constants now and is re-exported here so
+existing `from core.constants import internal_api_base` callers keep working.
+"""
+from src.constants import *  # noqa: F401,F403
+from src.constants import internal_api_base  # noqa: F401  (explicit: functions aren't covered by some linters' * checks)

core/database.py

@@ -29,8 +29,9 @@ class TimestampMixin:
     def updated_at(cls):
         return Column(DateTime, default=utcnow_naive, onupdate=utcnow_naive, nullable=False)
 
-# Get database URL from environment, default to SQLite
-DATABASE_URL = os.getenv("DATABASE_URL", "sqlite:///./data/app.db")
+# Get database URL from environment, default to SQLite in DATA_DIR
+from src.constants import DATA_DIR, AUTH_FILE, MEMORY_FILE, USER_PREFS_FILE, SETTINGS_FILE
+DATABASE_URL = os.getenv("DATABASE_URL", f"sqlite:///{DATA_DIR}/app.db")
 
 # Create engine
 engine = create_engine(
@@ -360,6 +361,24 @@ class ModelEndpoint(TimestampMixin, Base):
     # is the historical default. When non-null, the model picker only shows
     # the endpoint to that user (admins always see everything).
     owner = Column(String, nullable=True, index=True)
+    # Optional OAuth/session-backed credential row. Used by subscription-backed
+    # providers that need refresh tokens instead of a static API key.
+    provider_auth_id = Column(String, nullable=True, index=True)
+
+
+class ProviderAuthSession(TimestampMixin, Base):
+    """Encrypted OAuth/session credentials for refresh-aware model providers."""
+    __tablename__ = "provider_auth_sessions"
+
+    id = Column(String, primary_key=True, index=True)
+    provider = Column(String, nullable=False, index=True)
+    owner = Column(String, nullable=True, index=True)
+    label = Column(String, nullable=True)
+    base_url = Column(String, nullable=False)
+    access_token = Column(EncryptedText, nullable=True)
+    refresh_token = Column(EncryptedText, nullable=True)
+    last_refresh = Column(DateTime, nullable=True)
+    auth_mode = Column(String, nullable=True)
 
 class McpServer(TimestampMixin, Base):
     """Admin-configured MCP (Model Context Protocol) tool servers."""
@@ -669,6 +688,7 @@ def _migrate_add_last_message_at_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(sessions)")
@@ -694,10 +714,14 @@ def _migrate_add_last_message_at_column():
             "ON sessions(archived, last_message_at)"
         )
         conn.commit()
-        conn.close()
         logging.getLogger(__name__).info("Migrated: added + backfilled 'last_message_at' on sessions")
     except Exception as e:
         logging.getLogger(__name__).warning(f"last_message_at migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_document_archived_column():
     """Add `archived` to documents (soft-archive flag). Guarded + idempotent."""
@@ -705,6 +729,7 @@ def _migrate_add_document_archived_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(documents)")
@@ -713,9 +738,13 @@ def _migrate_add_document_archived_column():
             conn.execute("ALTER TABLE documents ADD COLUMN archived BOOLEAN DEFAULT 0")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'archived' to documents")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"documents.archived migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_owner_column():
@@ -724,6 +753,7 @@ def _migrate_add_owner_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(sessions)")
@@ -733,9 +763,13 @@ def _migrate_add_owner_column():
             conn.execute("CREATE INDEX IF NOT EXISTS ix_sessions_owner ON sessions(owner)")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'owner' column to sessions")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"Migration check failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_model_endpoints():
     """Recreate model_endpoints table if schema changed (url->base_url)."""
@@ -743,6 +777,7 @@ def _migrate_model_endpoints():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -751,9 +786,13 @@ def _migrate_model_endpoints():
             conn.execute("DROP TABLE IF EXISTS model_endpoints")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: dropped old model_endpoints table (schema change)")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"model_endpoints migration check failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_hidden_models_column():
     """Add hidden_models column to model_endpoints if it doesn't exist."""
@@ -761,6 +800,7 @@ def _migrate_add_hidden_models_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -769,9 +809,13 @@ def _migrate_add_hidden_models_column():
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN hidden_models TEXT")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'hidden_models' column to model_endpoints")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"hidden_models migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_model_endpoint_owner_column():
     """Add owner column to model_endpoints if it doesn't exist.
@@ -786,6 +830,7 @@ def _migrate_add_model_endpoint_owner_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -795,9 +840,38 @@ def _migrate_add_model_endpoint_owner_column():
             conn.execute("CREATE INDEX IF NOT EXISTS ix_model_endpoints_owner ON model_endpoints(owner)")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'owner' column + index to model_endpoints")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"model_endpoints.owner migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
+
+
+def _migrate_add_provider_auth_id_column():
+    """Add provider_auth_id column to model_endpoints if it doesn't exist."""
+    import sqlite3
+    db_path = DATABASE_URL.replace("sqlite:///", "")
+    if not os.path.exists(db_path):
+        return
+    conn = None
+    try:
+        conn = sqlite3.connect(db_path)
+        cursor = conn.execute("PRAGMA table_info(model_endpoints)")
+        columns = [row[1] for row in cursor.fetchall()]
+        if columns and "provider_auth_id" not in columns:
+            conn.execute("ALTER TABLE model_endpoints ADD COLUMN provider_auth_id VARCHAR")
+            conn.execute("CREATE INDEX IF NOT EXISTS ix_model_endpoints_provider_auth_id ON model_endpoints(provider_auth_id)")
+            conn.commit()
+            logging.getLogger(__name__).info("Migrated: added 'provider_auth_id' column + index to model_endpoints")
+    except Exception as e:
+        logging.getLogger(__name__).warning(f"model_endpoints.provider_auth_id migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_model_type_column():
@@ -806,6 +880,7 @@ def _migrate_add_model_type_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -814,9 +889,13 @@ def _migrate_add_model_type_column():
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN model_type TEXT DEFAULT 'llm'")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'model_type' column to model_endpoints")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"model_type migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_model_endpoint_refresh_columns():
     """Add endpoint classification / refresh policy columns if missing."""
@@ -824,6 +903,7 @@ def _migrate_add_model_endpoint_refresh_columns():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -837,9 +917,13 @@ def _migrate_add_model_endpoint_refresh_columns():
         if columns and "model_refresh_timeout" not in columns:
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN model_refresh_timeout INTEGER")
         conn.commit()
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"model_endpoints refresh-policy migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_task_run_model_column():
     """Add model column to task_runs if it doesn't exist (records which model ran)."""
@@ -847,6 +931,7 @@ def _migrate_add_task_run_model_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(task_runs)")
@@ -855,9 +940,13 @@ def _migrate_add_task_run_model_column():
             conn.execute("ALTER TABLE task_runs ADD COLUMN model TEXT")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'model' column to task_runs")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"task_runs model migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_supports_tools_column():
     """Add supports_tools column to model_endpoints if it doesn't exist."""
@@ -865,6 +954,7 @@ def _migrate_add_supports_tools_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -873,9 +963,13 @@ def _migrate_add_supports_tools_column():
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN supports_tools BOOLEAN")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'supports_tools' column to model_endpoints")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"supports_tools migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_cached_models_column():
@@ -884,6 +978,7 @@ def _migrate_add_cached_models_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -891,9 +986,13 @@ def _migrate_add_cached_models_column():
         if columns and "cached_models" not in columns:
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN cached_models TEXT")
             conn.commit()
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"cached_models migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_pinned_models_column():
     """Add pinned_models column to model_endpoints if it doesn't exist."""
@@ -901,6 +1000,7 @@ def _migrate_add_pinned_models_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(model_endpoints)")
@@ -909,9 +1009,13 @@ def _migrate_add_pinned_models_column():
             conn.execute("ALTER TABLE model_endpoints ADD COLUMN pinned_models TEXT")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'pinned_models' column to model_endpoints")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"pinned_models migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_notes_sort_order():
     """Add sort_order, image_url, repeat columns to notes if they don't exist."""
@@ -919,6 +1023,7 @@ def _migrate_add_notes_sort_order():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(notes)")
@@ -936,9 +1041,13 @@ def _migrate_add_notes_sort_order():
         if columns and "agent_session_id" not in columns:
             conn.execute("ALTER TABLE notes ADD COLUMN agent_session_id TEXT")
         conn.commit()
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"notes migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_mode_column():
     """Add mode column to sessions table if it doesn't exist."""
@@ -946,6 +1055,7 @@ def _migrate_add_mode_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(sessions)")
@@ -954,9 +1064,13 @@ def _migrate_add_mode_column():
             conn.execute("ALTER TABLE sessions ADD COLUMN mode TEXT")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'mode' column to sessions")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"Migration check for mode failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_folder_column():
     """Add folder column to sessions table if it doesn't exist."""
@@ -964,6 +1078,7 @@ def _migrate_add_folder_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(sessions)")
@@ -972,9 +1087,13 @@ def _migrate_add_folder_column():
             conn.execute("ALTER TABLE sessions ADD COLUMN folder TEXT")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'folder' column to sessions")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"Migration check for folder failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_token_columns():
     """Add cumulative token tracking columns to sessions table."""
@@ -982,6 +1101,7 @@ def _migrate_add_token_columns():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(sessions)")
@@ -991,9 +1111,13 @@ def _migrate_add_token_columns():
             conn.execute("ALTER TABLE sessions ADD COLUMN total_output_tokens INTEGER DEFAULT 0")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added token tracking columns to sessions")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"Migration check for token columns failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_owner_to_table(table_name: str, index_name: str):
     """Generic helper: add owner TEXT column + index to a table if missing."""
@@ -1001,6 +1125,7 @@ def _migrate_add_owner_to_table(table_name: str, index_name: str):
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute(f"PRAGMA table_info({table_name})")
@@ -1010,9 +1135,13 @@ def _migrate_add_owner_to_table(table_name: str, index_name: str):
             conn.execute(f"CREATE INDEX IF NOT EXISTS {index_name} ON {table_name}(owner)")
             conn.commit()
             logging.getLogger(__name__).info(f"Migrated: added 'owner' column to {table_name}")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"Migration owner column for {table_name} failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_add_multiuser_owner_columns():
     """Add owner column to memories, gallery_images, user_tools, comparisons."""
@@ -1037,6 +1166,7 @@ def _migrate_add_api_token_scopes_column():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         columns = [row[1] for row in conn.execute("PRAGMA table_info(api_tokens)").fetchall()]
@@ -1045,9 +1175,13 @@ def _migrate_add_api_token_scopes_column():
             conn.execute("UPDATE api_tokens SET scopes = 'chat' WHERE scopes IS NULL OR scopes = ''")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added scopes column to api_tokens")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"api_tokens.scopes migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def _migrate_assign_legacy_owner():
     """Assign all null-owner data to the first (admin) user.
@@ -1065,7 +1199,7 @@ def _migrate_assign_legacy_owner():
     # fell through to "first user" every time.
     auth_path = os.path.join(os.path.dirname(DATABASE_URL.replace("sqlite:///", "")), "auth.json")
     if not os.path.isabs(auth_path):
-        auth_path = os.path.join("data", "auth.json")
+        auth_path = AUTH_FILE
     admin_user = None
     try:
         with open(auth_path, "r", encoding="utf-8") as f:
@@ -1089,6 +1223,7 @@ def _migrate_assign_legacy_owner():
         return
 
     logger = logging.getLogger(__name__)
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         # Every table with an `owner` column. New tables added later will be
@@ -1113,12 +1248,16 @@ def _migrate_assign_legacy_owner():
             except Exception as e:
                 logger.warning(f"Legacy owner assignment for {table} failed: {e}")
         conn.commit()
-        conn.close()
     except Exception as e:
         logger.warning(f"Legacy owner migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
     # Also migrate memory.json
-    mem_path = os.path.join("data", "memory.json")
+    mem_path = MEMORY_FILE
     try:
         if os.path.exists(mem_path):
             with open(mem_path, "r", encoding="utf-8") as f:
@@ -1136,7 +1275,7 @@ def _migrate_assign_legacy_owner():
         logger.warning(f"memory.json legacy migration failed: {e}")
 
     # Also migrate user_prefs.json to per-user format
-    prefs_path = os.path.join("data", "user_prefs.json")
+    prefs_path = USER_PREFS_FILE
     try:
         if os.path.exists(prefs_path):
             with open(prefs_path, "r", encoding="utf-8") as f:
@@ -1458,7 +1597,11 @@ class CalendarCal(TimestampMixin, Base):
     owner = Column(String, nullable=True, index=True)
     name  = Column(String, nullable=False)
     color = Column(String, default="#5b8abf")
-    source = Column(String, default="local")  # "local" or "timetree"
+    source = Column(String, default="local")  # "local" or "caldav"
+    # UUID of the CalDAV account in user prefs that owns this calendar.
+    # NULL for local calendars and for CalDAV calendars created before
+    # multi-account support was added (treated as "use any configured account").
+    account_id = Column(String, nullable=True, index=True)
 
     events = relationship("CalendarEvent", back_populates="calendar", cascade="all, delete-orphan")
 
@@ -1526,7 +1669,7 @@ def _migrate_seed_email_account():
         import json as _json
         import uuid as _uuid
         from pathlib import Path
-        settings_file = Path("data/settings.json")
+        settings_file = Path(SETTINGS_FILE)
         if not settings_file.exists():
             return
         try:
@@ -1594,6 +1737,7 @@ def init_db():
     _migrate_add_model_type_column()
     _migrate_add_model_endpoint_refresh_columns()
     _migrate_add_model_endpoint_owner_column()
+    _migrate_add_provider_auth_id_column()
     _migrate_add_supports_tools_column()
     _migrate_add_task_run_model_column()
     _migrate_add_owner_column()
@@ -1622,9 +1766,105 @@ def init_db():
     _migrate_add_calendar_metadata()
     _migrate_add_calendar_is_utc()
     _migrate_add_calendar_origin()
+    _migrate_add_calendar_account_id()
+    _migrate_chat_messages_fts()
     _migrate_encrypt_email_passwords()
     _migrate_encrypt_signatures()
     _migrate_encrypt_endpoint_keys()
+    _migrate_backfill_task_folders()
+
+
+def _migrate_backfill_task_folders():
+    """Backfill folder='Tasks' on pre-existing task/research sessions.
+
+    Sessions created by the task scheduler (LLM tasks, action tasks, research
+    runs) now set folder='Tasks' at creation time.  This migration tags any
+    older sessions that predate that assignment.  Idempotent — only touches
+    rows where folder is NULL or empty and the title matches known prefixes.
+    """
+    try:
+        with engine.connect() as conn:
+            cols = [r[1] for r in conn.execute(text("PRAGMA table_info(sessions)"))]
+            if "folder" not in cols:
+                return
+            res = conn.execute(text(
+                "UPDATE sessions SET folder = 'Tasks' "
+                "WHERE (folder IS NULL OR folder = '') "
+                "AND (name LIKE '[Task] %' OR name LIKE '[Research] %')"
+            ))
+            conn.commit()
+            if res.rowcount:
+                logging.getLogger(__name__).info(
+                    f"Backfilled folder='Tasks' on {res.rowcount} task/research sessions")
+    except Exception as e:
+        logging.getLogger(__name__).warning(f"task folder backfill: {e}")
+
+
+def _migrate_chat_messages_fts():
+    """Create and backfill the session transcript FTS index for SQLite."""
+    if not DATABASE_URL.startswith("sqlite"):
+        return
+
+    db_path = DATABASE_URL.replace("sqlite:///", "")
+    if db_path == ":memory:":
+        return
+    conn = None
+    try:
+        conn = sqlite3.connect(db_path)
+        try:
+            conn.execute("CREATE VIRTUAL TABLE IF NOT EXISTS temp._odysseus_fts5_probe USING fts5(content)")
+            conn.execute("DROP TABLE IF EXISTS temp._odysseus_fts5_probe")
+        except Exception as e:
+            logging.getLogger(__name__).warning(f"chat_messages FTS migration skipped; FTS5 unavailable: {e}")
+            return
+
+        conn.executescript(
+            """
+            CREATE VIRTUAL TABLE IF NOT EXISTS chat_messages_fts USING fts5(
+                content,
+                message_id UNINDEXED,
+                session_id UNINDEXED,
+                role UNINDEXED
+            );
+
+            CREATE TRIGGER IF NOT EXISTS chat_messages_fts_ai
+            AFTER INSERT ON chat_messages BEGIN
+                INSERT INTO chat_messages_fts(content, message_id, session_id, role)
+                VALUES (COALESCE(new.content, ''), new.id, new.session_id, new.role);
+            END;
+
+            CREATE TRIGGER IF NOT EXISTS chat_messages_fts_ad
+            AFTER DELETE ON chat_messages BEGIN
+                DELETE FROM chat_messages_fts WHERE message_id = old.id;
+            END;
+
+            CREATE TRIGGER IF NOT EXISTS chat_messages_fts_au
+            AFTER UPDATE ON chat_messages BEGIN
+                DELETE FROM chat_messages_fts WHERE message_id = old.id;
+                INSERT INTO chat_messages_fts(content, message_id, session_id, role)
+                VALUES (COALESCE(new.content, ''), new.id, new.session_id, new.role);
+            END;
+            """
+        )
+        conn.execute(
+            """
+            INSERT INTO chat_messages_fts(content, message_id, session_id, role)
+            SELECT COALESCE(cm.content, ''), cm.id, cm.session_id, cm.role
+            FROM chat_messages cm
+            WHERE NOT EXISTS (
+                SELECT 1 FROM chat_messages_fts fts
+                WHERE fts.message_id = cm.id
+            )
+            """
+        )
+        conn.commit()
+    except Exception as e:
+        logging.getLogger(__name__).warning(f"chat_messages FTS migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_email_smtp_security():
@@ -1633,6 +1873,7 @@ def _migrate_add_email_smtp_security():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(email_accounts)")
@@ -1648,9 +1889,13 @@ def _migrate_add_email_smtp_security():
             )
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added smtp_security column to email_accounts")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"smtp_security migration skipped: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_encrypt_endpoint_keys():
@@ -1751,6 +1996,7 @@ def _migrate_add_calendar_is_utc():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(calendar_events)")
@@ -1759,9 +2005,13 @@ def _migrate_add_calendar_is_utc():
             conn.execute("ALTER TABLE calendar_events ADD COLUMN is_utc BOOLEAN DEFAULT 0 NOT NULL")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'is_utc' column to calendar_events")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"is_utc migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_calendar_origin():
@@ -1772,6 +2022,7 @@ def _migrate_add_calendar_origin():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(calendar_events)")
@@ -1781,9 +2032,39 @@ def _migrate_add_calendar_origin():
             conn.execute("CREATE INDEX IF NOT EXISTS ix_calendar_events_origin ON calendar_events(origin)")
             conn.commit()
             logging.getLogger(__name__).info("Migrated: added 'origin' column to calendar_events")
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"calendar_events.origin migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
+
+
+def _migrate_add_calendar_account_id():
+    """Add `account_id` to calendars so each CalDAV-backed calendar knows which
+    credential set (from caldav_accounts in user prefs) owns it. Idempotent."""
+    import sqlite3
+    db_path = DATABASE_URL.replace("sqlite:///", "")
+    if not os.path.exists(db_path):
+        return
+    conn = None
+    try:
+        conn = sqlite3.connect(db_path)
+        cursor = conn.execute("PRAGMA table_info(calendars)")
+        columns = [row[1] for row in cursor.fetchall()]
+        if columns and "account_id" not in columns:
+            conn.execute("ALTER TABLE calendars ADD COLUMN account_id TEXT")
+            conn.execute("CREATE INDEX IF NOT EXISTS ix_calendars_account_id ON calendars(account_id)")
+            conn.commit()
+            logging.getLogger(__name__).info("Migrated: added 'account_id' column to calendars")
+    except Exception as e:
+        logging.getLogger(__name__).warning(f"calendars.account_id migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 
 def _migrate_add_calendar_metadata():
@@ -1792,6 +2073,7 @@ def _migrate_add_calendar_metadata():
     db_path = DATABASE_URL.replace("sqlite:///", "")
     if not os.path.exists(db_path):
         return
+    conn = None
     try:
         conn = sqlite3.connect(db_path)
         cursor = conn.execute("PRAGMA table_info(calendar_events)")
@@ -1803,9 +2085,13 @@ def _migrate_add_calendar_metadata():
         if columns and "last_pinged" not in columns:
             conn.execute("ALTER TABLE calendar_events ADD COLUMN last_pinged DATETIME")
         conn.commit()
-        conn.close()
     except Exception as e:
         logging.getLogger(__name__).warning(f"calendar_events migration failed: {e}")
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
 
 def get_db():
     """

core/middleware.py

@@ -17,6 +17,15 @@ INTERNAL_TOOL_TOKEN = os.environ.get("ODYSSEUS_INTERNAL_TOKEN") or secrets.token
 INTERNAL_TOOL_HEADER = "X-Odysseus-Internal-Token"
 
 
+def is_cors_preflight(method: str, headers) -> bool:
+    """True for a genuine CORS preflight: an OPTIONS request carrying the
+    Access-Control-Request-Method header. Such requests are credential-less by
+    design and must reach CORSMiddleware to be answered -- gating them on auth
+    401s the preflight and breaks every cross-origin browser/WebView client.
+    Pure so it can be unit-tested without standing up the app."""
+    return method == "OPTIONS" and "access-control-request-method" in headers
+
+
 def require_admin(request: Request):
     """Raise 403 if the current user isn't an admin.
     Allows access when auth is explicitly disabled, or when the request carries
@@ -58,11 +67,22 @@ class SecurityHeadersMiddleware(BaseHTTPMiddleware):
 
         # Tool render endpoints are served inside iframes — allow framing by self
         is_tool_render = path.startswith("/api/tools/") and path.endswith("/render")
+        # PDF previews are embedded by the in-app document library. Keep the
+        # exception route-scoped so normal app pages remain unframeable.
+        is_document_pdf_preview = path.startswith("/api/document/") and path.endswith("/render-pdf")
         # Visual report pages are self-contained HTML — need inline scripts + external images
         is_report = path.startswith("/api/research/report/")
 
         response.headers["X-Content-Type-Options"] = "nosniff"
         response.headers["Referrer-Policy"] = "no-referrer"
+        response.headers["Permissions-Policy"] = "camera=(), microphone=(self), geolocation=()"
+
+        is_https = (
+            request.url.scheme == "https"
+            or request.headers.get("X-Forwarded-Proto") == "https"
+        )
+        if is_https:
+            response.headers["Strict-Transport-Security"] = "max-age=31536000; includeSubDomains"
 
         if is_report:
             response.headers["Content-Security-Policy"] = (
@@ -79,6 +99,12 @@ class SecurityHeadersMiddleware(BaseHTTPMiddleware):
             # sandbox="allow-scripts" attribute provides isolation.
             # Don't overwrite the route's own restrictive CSP either.
             pass
+        elif is_document_pdf_preview:
+            response.headers["X-Frame-Options"] = "SAMEORIGIN"
+            response.headers["Content-Security-Policy"] = (
+                "default-src 'none'; "
+                "frame-ancestors 'self'"
+            )
         else:
             response.headers["X-Frame-Options"] = "DENY"
             # NOTE: `style-src 'unsafe-inline'` is intentionally retained.

core/models.py

@@ -11,14 +11,24 @@ from typing import Dict, List, Any, Optional, TYPE_CHECKING
 if TYPE_CHECKING:
     from .session_manager import SessionManager
 
-# Module-level session manager reference (set at app startup)
-_session_manager: Optional["SessionManager"] = None
+# Module-level session manager singleton (single source of truth)
+_SESSION_MANAGER_INSTANCE: Optional["SessionManager"] = None
 
 
-def set_session_manager(manager: "SessionManager"):
-    """Set the global session manager reference."""
-    global _session_manager
-    _session_manager = manager
+def set_session_manager_instance(manager: "SessionManager"):
+    """Set the global SessionManager singleton."""
+    global _SESSION_MANAGER_INSTANCE
+    _SESSION_MANAGER_INSTANCE = manager
+
+
+def get_session_manager_instance() -> Optional["SessionManager"]:
+    """Get the global SessionManager singleton."""
+    return _SESSION_MANAGER_INSTANCE
+
+
+# Keep legacy name for backward compatibility
+set_session_manager = set_session_manager_instance
+get_session_manager = get_session_manager_instance
 
 
 @dataclass
@@ -42,7 +52,17 @@ class ChatMessage:
 
 @dataclass
 class Session:
-    """A chat session — pure data container."""
+    """A chat session — pure data container.
+
+    ``.history`` is the authoritative mutable message list. Callers may
+    read, append, pop, or reassign it directly — these changes take
+    effect immediately. ``_history`` remains a compatibility alias that
+    always resolves to the authoritative ``history`` list.
+
+    Each session gets its own unique history list at construction time
+    (the dataclass default is never shared between instances).
+    """
+
     id: str
     name: str
     endpoint_url: str
@@ -56,24 +76,35 @@ class Session:
     message_count: int = 0
 
     def __post_init__(self):
-        if self.history is None:
-            self.history = []
         if self.headers is None:
             self.headers = {}
+        # Ensure each session gets its OWN list (not the shared dataclass default)
+        if self.history is None:
+            self.history = []
+
+    @property
+    def _history(self) -> List[ChatMessage]:
+        """Compatibility alias for callers that still reference ``_history``."""
+        return self.history
+
+    @_history.setter
+    def _history(self, messages: List[ChatMessage]):
+        self.history = messages
 
     def add_message(self, message: ChatMessage):
         """
         Add a message to this session.
 
-        Delegates to SessionManager for persistence if available,
-        otherwise just appends to history.
+        Appends to the authoritative history list and increments
+        message_count. Delegates to SessionManager for persistence
+        if available.
         """
         self.history.append(message)
         self.message_count = len(self.history)
 
         # Delegate to session manager for persistence
-        if _session_manager:
-            _session_manager._persist_message(self.id, message)
+        if _SESSION_MANAGER_INSTANCE:
+            _SESSION_MANAGER_INSTANCE._persist_message(self.id, message)
 
     def get_context_messages(self) -> List[Dict[str, Any]]:
         """Get messages in format for LLM API.
@@ -94,3 +125,7 @@ class Session:
     def get(self, key: str, default=None):
         """Dict-like access for compatibility."""
         return getattr(self, key, default)
+
+    def __getitem__(self, key: str):
+        """Allow session['field'] syntax."""
+        return getattr(self, key)

core/platform_compat.py

@@ -18,10 +18,22 @@ import ntpath
 import shutil
 import subprocess
 from pathlib import Path
+import sys
 from typing import List, Optional
+import platform
 
 IS_WINDOWS = os.name == "nt"
 IS_POSIX = not IS_WINDOWS
+# Allows APFEL support and ARM-native binary recommendations on Apple Silicon Macs.
+IS_APPLE_SILICON = (
+    IS_POSIX
+    and platform.system() == "Darwin"
+    and platform.machine().lower()
+    in {
+        "arm64",
+        "aarch64",
+    }
+)
 
 
 # ── File permissions ────────────────────────────────────────────────────────
@@ -53,9 +65,8 @@ def detached_popen_kwargs() -> dict:
     and is detached from any console.
     """
     if IS_WINDOWS:
-        flags = (
-            getattr(subprocess, "CREATE_NEW_PROCESS_GROUP", 0x00000200)
-            | getattr(subprocess, "DETACHED_PROCESS", 0x00000008)
+        flags = getattr(subprocess, "CREATE_NEW_PROCESS_GROUP", 0x00000200) | getattr(
+            subprocess, "DETACHED_PROCESS", 0x00000008
         )
         return {"creationflags": flags}
     return {"start_new_session": True}
@@ -150,6 +161,29 @@ _WINDOWS_BASH_RELATIVE_PATHS = (
     ("usr", "bin", "bash.exe"),
 )
 
+# Paths to add to the remote SSH probe command to find tools like nvidia-smi that may not be on PATH.
+_SSH_PATH_MEMBERS = (
+    "/usr/bin",
+    "/usr/local/bin",
+    "/usr/local/cuda/bin",
+    "/usr/lib/wsl/lib"
+)
+# Fallback locations for nvidia-smi on WSL and other Linux distros where it may not be on PATH.
+NVIDIA_PATH_CANDIDATES = (
+    "/usr/bin/nvidia-smi",
+    "/usr/local/bin/nvidia-smi",
+    "/usr/local/cuda/bin/nvidia-smi",
+    "/usr/lib/wsl/lib/nvidia-smi",
+)
+
+
+def _ssh_path_override() -> str:
+    """Build the PATH export snippet used for remote SSH shell probes."""
+    return f"export PATH=\"$PATH:{':'.join(_SSH_PATH_MEMBERS)}\"; "
+
+
+SSH_PATH_OVERRIDE = _ssh_path_override()
+
 
 def _windows_bash_fallbacks() -> List[str]:
     roots: List[str] = []
@@ -157,6 +191,8 @@ def _windows_bash_fallbacks() -> List[str]:
         base = os.environ.get(env_name)
         if base:
             roots.append(ntpath.join(base, "Git"))
+            if env_name == "LocalAppData":
+                roots.append(ntpath.join(base, "Programs", "Git"))
     roots.extend(_WINDOWS_BASH_DEFAULT_ROOTS)
 
     paths: List[str] = []
@@ -180,6 +216,21 @@ def _is_windows_bash_stub(path: str) -> bool:
     )
 
 
+def git_bash_path(path: str | Path) -> str:
+    """Convert a path to POSIX style suitable for Git Bash on Windows.
+
+    Transforms drive letters (e.g., 'C:\\path') to POSIX '/c/path',
+    and uses forward slashes.
+    """
+    p = Path(path)
+    p_str = p.as_posix()
+    if IS_WINDOWS and len(p_str) >= 2 and p_str[1] == ":":
+        drive = p_str[0].lower()
+        return f"/{drive}{p_str[2:]}"
+    return p_str
+
+
+
 def find_bash() -> Optional[str]:
     """Locate a real ``bash`` interpreter, or None.
 
@@ -242,3 +293,160 @@ def run_script_argv(script_path) -> List[str]:
         comspec = os.environ.get("ComSpec", "cmd.exe")
         return [comspec, "/c", str(script_path)]
     return ["sh", str(script_path)]
+
+
+def is_wsl() -> bool:
+    """True if running inside Windows Subsystem for Linux (WSL)."""
+    import sys
+    if sys.platform.startswith("linux") or os.name == "posix":
+        try:
+            with open("/proc/version", "r", encoding="utf-8", errors="ignore") as f:
+                if "microsoft" in f.read().lower():
+                    return True
+        except Exception:
+            pass
+    return False
+
+
+def translate_path(path_str: str) -> str:
+    """Translate a path (possibly a Windows path) to the current OS format.
+
+    Particularly handles Windows paths (e.g. C:\\foo or C:/foo) when running
+    under WSL, translating them to /mnt/c/foo.
+    Also handles standard path normalization to avoid string breakages.
+    """
+    if not path_str:
+        return path_str
+
+    if is_wsl():
+        path_str = path_str.replace("\\", "/")
+        import re
+        m = re.match(r"^([a-zA-Z]):(.*)", path_str)
+        if m:
+            drive = m.group(1).lower()
+            rest = m.group(2)
+            if not rest.startswith("/"):
+                rest = "/" + rest
+            return f"/mnt/{drive}{rest}"
+
+    try:
+        return str(Path(path_str).resolve())
+    except Exception:
+        return path_str
+
+
+def get_wsl_windows_user_profile() -> Optional[str]:
+    """Retrieve the Windows host User Profile path from inside WSL."""
+    if not is_wsl():
+        return None
+    try:
+        r = run_wsl_windows_powershell("Write-Output $env:USERPROFILE", timeout=5)
+        if r.returncode == 0 and r.stdout.strip():
+            return translate_path(r.stdout.strip())
+    except Exception:
+        pass
+
+    try:
+        users_dir = "/mnt/c/Users"
+        if os.path.isdir(users_dir):
+            for entry in os.listdir(users_dir):
+                if entry not in ("All Users", "Default", "Default User", "desktop.ini", "Public"):
+                    path = os.path.join(users_dir, entry)
+                    if os.path.isdir(path):
+                        return path
+    except Exception:
+        pass
+    return None
+
+
+def _ssh_exec_argv(
+    remote: str,
+    ssh_port: str | None,
+    *,
+    remote_cmd: str | None = None,
+    connect_timeout: int | None = None,
+    strict_host_key_checking: bool | None = None,
+) -> list[str]:
+    """Build a consistent ssh argv for remote command execution."""
+    remote_value = str(remote or "").strip()
+    remote_host = remote_value.rsplit("@", 1)[-1]
+    if not remote_value or remote_value.startswith("-") or not remote_host or remote_host.startswith("-"):
+        raise ValueError("Invalid SSH remote host")
+    argv = ["ssh"]
+    if connect_timeout is not None:
+        argv.extend(["-o", f"ConnectTimeout={int(connect_timeout)}"])
+    if strict_host_key_checking is not None:
+        argv.extend(
+            [
+                "-o",
+                "StrictHostKeyChecking=yes"
+                if strict_host_key_checking
+                else "StrictHostKeyChecking=no",
+            ]
+        )
+    if ssh_port and ssh_port != "22":
+        argv.extend(["-p", str(ssh_port)])
+    argv.append(remote)
+    if remote_cmd is not None:
+        argv.append(remote_cmd)
+    return argv
+
+
+def run_ssh_command(
+    remote: str,
+    ssh_port: str | None,
+    remote_cmd: str,
+    *,
+    timeout: float,
+    connect_timeout: int | None = None,
+    strict_host_key_checking: bool | None = None,
+    text: bool = True,
+) -> subprocess.CompletedProcess:
+    """Run an ssh command with centralized timeout and stderr/stdout capture."""
+    return subprocess.run(
+        _ssh_exec_argv(
+            remote,
+            ssh_port,
+            remote_cmd=remote_cmd,
+            connect_timeout=connect_timeout,
+            strict_host_key_checking=strict_host_key_checking,
+        ),
+        timeout=timeout,
+        capture_output=True,
+        text=text,
+    )
+
+
+def _windows_powershell_argv(
+    command: str,
+    *,
+    no_profile: bool = True,
+    non_interactive: bool = True,
+) -> List[str]:
+    argv: List[str] = ["powershell.exe"]
+    if no_profile:
+        argv.append("-NoProfile")
+    if non_interactive:
+        argv.append("-NonInteractive")
+    argv.extend(["-Command", command])
+    return argv
+
+
+def run_wsl_windows_powershell(
+    command: str,
+    *,
+    timeout: float = 5,
+) -> subprocess.CompletedProcess[str]:
+    """Run a PowerShell command on the Windows host from WSL.
+
+    Raises ``RuntimeError`` when called outside WSL.
+    """
+
+    if not is_wsl():
+        raise RuntimeError("run_wsl_windows_powershell is only supported in WSL")
+    return subprocess.run(
+        _windows_powershell_argv(command),
+        capture_output=True,
+        text=True,
+        timeout=timeout,
+    )

core/session_manager.py

@@ -14,9 +14,12 @@ import logging
 from datetime import datetime, timezone, timedelta
 from typing import Dict, Optional
 
-from .database import Session as DbSession, ChatMessage as DbChatMessage, Document as DbDocument, SessionLocal
+from .database import Session as DbSession, ChatMessage as DbChatMessage, Document as DbDocument, SessionLocal, utcnow_naive
 from .models import Session, ChatMessage
 
+# Re-export singleton accessors from models for convenience
+from .models import set_session_manager_instance, get_session_manager_instance
+
 logger = logging.getLogger(__name__)
 
 
@@ -188,12 +191,17 @@ class SessionManager:
         """
         Add a message to a session and persist to database.
 
+        Updates the authoritative history list and persists through this
+        manager directly so tests and temporary managers do not depend on the
+        process-wide session-manager singleton.
+
         Args:
             session_id: Session ID
             message: ChatMessage to add
         """
         session = self.get_session(session_id)
         session.history.append(message)
+        session._history = session.history
         session.message_count = len(session.history)
 
         self._persist_message(session_id, message)
@@ -232,7 +240,10 @@ class SessionManager:
             )
             db.add(db_message)
 
-            db_session.message_count = len(self.sessions.get(session_id, {}).history) if session_id in self.sessions else 0
+            if session_id in self.sessions:
+                db_session.message_count = len(self.sessions[session_id].history)
+            else:
+                db_session.message_count = 0
             _now = datetime.now(timezone.utc)
             db_session.last_accessed = _now
             # Clean "last conversation" timestamp — only bumped here on a
@@ -283,6 +294,7 @@ class SessionManager:
 
             # Update in-memory
             session.history = session.history[:keep_count]
+            session._history = session.history
 
             logger.info(f"Truncated session {session_id} to {keep_count} messages")
             return True
@@ -333,6 +345,7 @@ class SessionManager:
 
             db.commit()
             session.history = list(messages)
+            session._history = session.history
             session.message_count = len(messages)
             logger.info("Replaced session %s history with %d messages", session_id, len(messages))
             return True
@@ -608,24 +621,52 @@ class SessionManager:
     def save_sessions(self):
         """No-op for DB compatibility."""
 
+    def ensure_task_session(self, session_id: str, name: str, endpoint_url: str, model: str, owner: str = None, task: object = None) -> Session:
+        """Create a task session if it doesn't exist, or return the existing one.
+
+        Unlike create_session, this checks the cache first and does NOT
+        overwrite an existing in-memory session. The task scheduler must
+        use this instead of direct dict assignment.
+        """
+        if session_id in self.sessions:
+            return self.sessions[session_id]
+
+        session = self.create_session(session_id, name, endpoint_url, model, owner=owner)
+        if task is not None:
+            task.session_id = session_id
+        return session
+
     # ------------------------------------------------------------------
     # Cleanup
     # ------------------------------------------------------------------
 
-    def cleanup_empty_sessions(self, auto_archive_days: int = 30) -> dict:
-        """Clean up empty and old sessions."""
+    def cleanup_empty_sessions(self, auto_archive_days: int = 30, min_age_hours: int = 1) -> dict:
+        """Clean up empty and old sessions.
+
+        Args:
+            auto_archive_days: Age in days before non-important sessions are archived.
+            min_age_hours: Minimum age in hours before an empty session can be deleted.
+                          Prevents deleting sessions that were just created.
+        """
         db = SessionLocal()
         stats = {'deleted_empty': 0, 'archived_old': 0, 'total_checked': 0}
 
         try:
             all_sessions = db.query(DbSession).all()
-            cutoff_date = datetime.now(timezone.utc) - timedelta(days=auto_archive_days)
+            cutoff_date = utcnow_naive() - timedelta(days=auto_archive_days)
+            min_age = utcnow_naive() - timedelta(hours=min_age_hours)
 
             for db_session in all_sessions:
                 stats['total_checked'] += 1
 
-                # Delete empty sessions
+                # Delete empty sessions only if older than min_age_hours
                 if db_session.message_count == 0:
+                    if db_session.created_at is not None:
+                        created = db_session.created_at
+                        if created.tzinfo is None:
+                            created = created.replace(tzinfo=timezone.utc)
+                        if created > min_age:
+                            continue  # Too young to delete
                     if db_session.id in self.sessions:
                         del self.sessions[db_session.id]
                     db.delete(db_session)

docker-compose.gpu-amd.yml

@@ -52,12 +52,14 @@ services:
       - SECURE_COOKIES=${SECURE_COOKIES:-false}
       - EMBEDDING_URL=${EMBEDDING_URL:-}
       - EMBEDDING_MODEL=${EMBEDDING_MODEL:-}
+      - EMBEDDING_API_KEY=${EMBEDDING_API_KEY:-}
       - FASTEMBED_MODEL=${FASTEMBED_MODEL:-sentence-transformers/all-MiniLM-L6-v2}
       - FASTEMBED_CACHE_PATH=${FASTEMBED_CACHE_PATH:-}
       - CLEANUP_INTERVAL_HOURS=${CLEANUP_INTERVAL_HOURS:-24}
       - ODYSSEUS_INPROCESS_POLLERS=${ODYSSEUS_INPROCESS_POLLERS:-1}
       - ODYSSEUS_INPROCESS_TASKS=${ODYSSEUS_INPROCESS_TASKS:-1}
       - ODYSSEUS_SCRIPT_HOST=${ODYSSEUS_SCRIPT_HOST:-localhost}
+      - ODYSSEUS_CHAT_UPLOAD_MAX_BYTES=${ODYSSEUS_CHAT_UPLOAD_MAX_BYTES:-10485760}
       - DATA_BRAVE_API_KEY=${DATA_BRAVE_API_KEY:-}
       - GOOGLE_API_KEY=${GOOGLE_API_KEY:-}
       - GOOGLE_PSE_CX=${GOOGLE_PSE_CX:-}

docker-compose.gpu-nvidia.yml

@@ -51,12 +51,14 @@ services:
       - SECURE_COOKIES=${SECURE_COOKIES:-false}
       - EMBEDDING_URL=${EMBEDDING_URL:-}
       - EMBEDDING_MODEL=${EMBEDDING_MODEL:-}
+      - EMBEDDING_API_KEY=${EMBEDDING_API_KEY:-}
       - FASTEMBED_MODEL=${FASTEMBED_MODEL:-sentence-transformers/all-MiniLM-L6-v2}
       - FASTEMBED_CACHE_PATH=${FASTEMBED_CACHE_PATH:-}
       - CLEANUP_INTERVAL_HOURS=${CLEANUP_INTERVAL_HOURS:-24}
       - ODYSSEUS_INPROCESS_POLLERS=${ODYSSEUS_INPROCESS_POLLERS:-1}
       - ODYSSEUS_INPROCESS_TASKS=${ODYSSEUS_INPROCESS_TASKS:-1}
       - ODYSSEUS_SCRIPT_HOST=${ODYSSEUS_SCRIPT_HOST:-localhost}
+      - ODYSSEUS_CHAT_UPLOAD_MAX_BYTES=${ODYSSEUS_CHAT_UPLOAD_MAX_BYTES:-10485760}
       - DATA_BRAVE_API_KEY=${DATA_BRAVE_API_KEY:-}
       - GOOGLE_API_KEY=${GOOGLE_API_KEY:-}
       - GOOGLE_PSE_CX=${GOOGLE_PSE_CX:-}

docker-compose.yml

@@ -40,12 +40,14 @@ services:
       - SECURE_COOKIES=${SECURE_COOKIES:-false}
       - EMBEDDING_URL=${EMBEDDING_URL:-}
       - EMBEDDING_MODEL=${EMBEDDING_MODEL:-}
+      - EMBEDDING_API_KEY=${EMBEDDING_API_KEY:-}
       - FASTEMBED_MODEL=${FASTEMBED_MODEL:-sentence-transformers/all-MiniLM-L6-v2}
       - FASTEMBED_CACHE_PATH=${FASTEMBED_CACHE_PATH:-}
       - CLEANUP_INTERVAL_HOURS=${CLEANUP_INTERVAL_HOURS:-24}
       - ODYSSEUS_INPROCESS_POLLERS=${ODYSSEUS_INPROCESS_POLLERS:-1}
       - ODYSSEUS_INPROCESS_TASKS=${ODYSSEUS_INPROCESS_TASKS:-1}
       - ODYSSEUS_SCRIPT_HOST=${ODYSSEUS_SCRIPT_HOST:-localhost}
+      - ODYSSEUS_CHAT_UPLOAD_MAX_BYTES=${ODYSSEUS_CHAT_UPLOAD_MAX_BYTES:-10485760}
       - DATA_BRAVE_API_KEY=${DATA_BRAVE_API_KEY:-}
       - GOOGLE_API_KEY=${GOOGLE_API_KEY:-}
       - GOOGLE_PSE_CX=${GOOGLE_PSE_CX:-}

docs/email-outlook.md

@@ -0,0 +1,17 @@
+# Outlook / Office 365 email accounts
+
+Odysseus email accounts currently use IMAP and SMTP with username/password
+authentication. That works for providers that still allow app passwords or
+mailbox passwords for IMAP/SMTP.
+
+Microsoft disables basic authentication for Outlook and Microsoft 365 in most
+modern accounts and tenants. If you try to add an Outlook account with a normal
+password, Microsoft may return errors such as:
+
+- `IMAP: AUTHENTICATE failed`
+- `SMTP: 535 5.7.139 Authentication unsuccessful, basic authentication is disabled`
+
+This is expected. Odysseus does not support Microsoft OAuth or Graph Mail yet,
+so Outlook / Office 365 accounts cannot currently be added through the password
+form. Use another email provider with app-password support, or track the future
+Microsoft Graph OAuth integration.

docs/index.html

@@ -25,9 +25,16 @@
     --radius: 8px;
   }
   * { box-sizing: border-box; }
-  html { scroll-behavior: smooth; scroll-snap-type: y proximity; scroll-padding-top: 60px; }
-  /* Each section is a full-viewport "page" with its content centered, so only
-     one shows at a time and the snap is obvious. */
+  html { scroll-behavior: smooth; scroll-padding-top: 60px; }
+  /* REMOVED: "scroll-snap-type: y proximity"
+     The idea was: >>Each section is a full-viewport "page" with its content centered,
+     so only one shows at a time and the snap is obvious.<<
+
+     PROBLEM: sections easily grow taller than 100vh IRL
+     This cause forced jumps mid-read. It's intrusive UX.
+     The landing-page is not a PowerPoint presentation!
+
+     Preserved: CSS snap-points to avoid destroying code meta-data*/
   .hero, section {
     scroll-snap-align: start; min-height: 100vh;
     display: flex; flex-direction: column; justify-content: center;

docs/security-ci.md

@@ -0,0 +1,102 @@
+# Security CI guide
+
+This project runs a set of automated security checks on every pull request and
+on every push to `main`. This page explains what each one does, whether it can
+block a merge, and the few one-time settings you should turn on to get the full
+benefit.
+
+## What runs, and why
+
+Each check lives in its own file under `.github/workflows/`. They run
+automatically; you do not start them.
+
+| Check | What it protects against | Blocks a merge? |
+|---|---|---|
+| **Secret scan** (gitleaks) | An API key, token, or password being committed by mistake or on purpose | Yes |
+| **Workflow security** (actionlint + zizmor) | A broken or insecure automation file that could leak the repo's access token | Yes |
+| **Dependency review** | A pull request that adds a software library with a known security hole | Yes |
+| **pip-audit** | Known security holes in the Python libraries already used | No (advisory) |
+| **Container scan: hadolint** | Mistakes and insecure patterns in the `Dockerfile` | Yes |
+| **Container scan: Trivy** | Known security holes in the Docker image | No (advisory) |
+| **CodeQL** | Real bugs in the app's own code: injection, auth mistakes, path traversal | No (advisory) |
+
+"Blocks a merge" means a red X appears on the pull request and, once you enable
+the setting below, the **Merge** button is disabled until it is fixed.
+
+"Advisory" means it reports problems into the repository's **Security** tab so
+you can review them on your own schedule, but it never stops a merge. These are
+advisory on purpose: they often flag long-standing issues in other people's
+libraries, not something a given pull request introduced.
+
+## Where results appear
+
+- **Checks tab of a pull request**: the pass/fail of each check. A green tick is
+  good; a red X needs attention.
+- **Security tab of the repository**: detailed findings from the advisory
+  scanners (Trivy and CodeQL). This is your dashboard.
+
+## If a check fails
+
+- **Secret scan failed**: a real credential may have been committed. Treat it as
+  leaked: rotate (regenerate) that key or token immediately, then remove it from
+  the file. Do not just delete the commit; assume it was seen.
+- **Dependency review failed**: the pull request adds a library with a known
+  vulnerability. Ask the contributor to use a patched version, or decline the
+  change.
+- **hadolint / workflow security failed**: the contributor changed the
+  `Dockerfile` or an automation file in a way the linter rejects. Ask them to
+  address the message shown in the failed check.
+
+## One-time settings to turn on
+
+These two settings unlock the full value. You only do them once.
+
+### 1. Require the blocking checks before merging
+
+This makes the **Merge** button refuse to work until the gating checks pass.
+
+1. Go to the repository on GitHub.
+2. Click **Settings** (top right of the repo).
+3. In the left sidebar, click **Branches**.
+4. Under **Branch protection rules**, click **Add branch ruleset** (or **Add
+   rule**), and set the branch name pattern to `dev` (this is the branch all
+   pull requests target; `main` is fast-forwarded at releases).
+5. Enable **Require status checks to pass before merging**.
+6. In the search box that appears, add these checks by name:
+   - `Python syntax (compileall)`
+   - `JS syntax (node --check)`
+   - `gitleaks`
+   - `actionlint`
+   - `zizmor (Actions SAST)`
+   - `hadolint (Dockerfile lint)`
+   - `dependency-review (PR gate)`
+
+   The first two come from the correctness CI (`ci.yml`); the rest are this
+   security suite. Leave pytest, pip-audit, Trivy, and CodeQL unchecked so they
+   stay advisory.
+7. Also enable **Require a pull request before merging** and **Require review
+   from Code Owners** (this uses the `.github/CODEOWNERS` file so every change
+   needs your sign-off).
+8. Click **Create** / **Save changes**.
+
+Note: a check name only appears in the list after it has run at least once, so
+let the workflows run on one pull request first, then add them here.
+
+### 2. Turn on the Security tab features
+
+1. **Settings -> Code security** (or **Code security and analysis**).
+2. Turn on **Dependency graph** (usually on by default for public repos) -- this
+   powers Dependency review and Dependabot.
+3. Turn on **Dependabot alerts** and **Dependabot security updates**.
+4. Under **Code scanning**, you have two ways to scan the app code with CodeQL:
+   - The included `codeql.yml` workflow already scans `main` and runs weekly.
+   - To also scan **pull requests** (recommended, since most contributions come
+     from forks), click **Set up -> Default** under Code scanning. GitHub then
+     runs CodeQL on pull requests for you, with no token limitations.
+
+## Keeping it current
+
+`.github/dependabot.yml` opens small weekly pull requests to update Python and
+npm packages, the Docker base image, and the pinned automation actions
+themselves. Review and merge those like any other pull request; they keep the
+project patched without manual tracking.

integrations/claude/skills/odysseus/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: odysseus
-description: Use when the user asks Claude Code to read or write Odysseus data (todos, email, calendar, memory, documents) through the scoped Claude Agent API. Requires ODYSSEUS_URL and ODYSSEUS_API_TOKEN.
+description: Use when the user asks Claude Code to read or write Odysseus data (todos, email, calendar, memory, documents) or to launch/monitor/stop a Cookbook model-serve task through the scoped Claude Agent API. Requires ODYSSEUS_URL and ODYSSEUS_API_TOKEN.
 ---
 
 # Odysseus
@@ -105,6 +105,49 @@ python3 ~/.claude/skills/odysseus/scripts/odysseus_api.py POST /api/codex/memory
 - `POST /api/codex/emails/draft` — body matches `SendEmailRequest` (`to`, `cc`, `bcc`, `subject`, `body`, `body_html`, `attachments`, `account_id`, `in_reply_to`, `references`). Requires `email:draft` (or `email:send`).
 - `POST /api/codex/emails/send` — same body. Requires `email:send`. Never send without explicit user instruction.
 
+## Cookbook serve (debug a failing model launch)
+
+The Cookbook surface lets you reproduce what a human would do in Odysseus → Cookbook: read which serves are running, tail their tmux output to see why they crashed, edit the launch command, relaunch, kill a stuck one. Use this when the user is debugging a model server that won't come up (compute-capability errors, OOM, missing kernels, wrong attention backend, etc.).
+
+- `GET /api/codex/cookbook/tasks` — list active serve/download/install tasks (sessionId, type, status, repo_id, remoteHost, payload._cmd). Requires `cookbook:read`.
+- `GET /api/codex/cookbook/servers` — list configured servers (name, host, port, env type + path, model dirs). Requires `cookbook:read`.
+- `GET /api/codex/cookbook/cached?host=<NAME>` — list models already cached on the named server (HF cache + Ollama + extra modelDirs). Call BEFORE `serve` to see what's already on disk. Requires `cookbook:read`.
+- `GET /api/codex/cookbook/presets` — list saved serve presets (model + host + port + cmd). The user's saved preset usually has a working cmd — try `preset NAME` before composing your own. Requires `cookbook:read`.
+- `GET /api/codex/cookbook/output/{session_id}?tail=400` — read the last N lines of the task's persistent log file (preferred) or tmux pane (fallback). The log file persists across vllm crashes, so this returns the actual Python traceback even after the bash prompt + neofetch banner overwrites the pane. Default tail=400. Requires `cookbook:read`.
+- `POST /api/codex/cookbook/serve` — launch a serve task. Body matches `ServeRequest`: `{ repo_id, cmd, remote_host?, ssh_port?, env_prefix?, gpus?, platform? }`. The `cmd` is validated: leading binary must be `vllm`/`python3`/`sglang`/`llama-server`/`ollama`/`node`/`npx`. NEVER prefix with `cd …`, `source …`, or chain with `&&`/`||`/`;`/`$(...)` — the validator rejects shell metacharacters. The venv activation (`env_prefix`) is added automatically from the host's saved settings, so pass the bare binary + args. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/preset/{name}` — launch a saved preset by name. Reuses the working cmd + host the user already saved. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/adopt` — register an externally-launched tmux session into cookbook tracking. Body: `{ tmux_session, model, host?, port? }`. Use this when serve_model rejected a cmd and you fell back to direct ssh+tmux — without adoption, the session is invisible to the UI. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/stop/{session_id}` — kill the tmux session for that task. Requires `cookbook:launch`.
+
+```bash
+# Survey what's running
+python3 ~/.claude/skills/odysseus/scripts/odysseus_api.py cookbook tasks
+
+# Tail the failing one (sessionId from `cookbook tasks`)
+python3 ~/.claude/skills/odysseus/scripts/odysseus_api.py cookbook output serve-abc12345 400
+
+# Stop the previous attempt before you try a new flag set
+python3 ~/.claude/skills/odysseus/scripts/odysseus_api.py cookbook stop serve-abc12345
+
+# Relaunch with new flags. cmd MUST begin with one of the allowlisted binaries.
+python3 ~/.claude/skills/odysseus/scripts/odysseus_api.py cookbook serve \
+  /mnt/HADES/models/Qwen3.5-397B-A17B-AWQ \
+  "vllm serve /mnt/HADES/models/Qwen3.5-397B-A17B-AWQ --host 0.0.0.0 --port 8001 --tensor-parallel-size 8 --max-model-len 262144 --gpu-memory-utilization 0.90 --dtype auto --max-num-seqs 8 --trust-remote-code --enable-expert-parallel --enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser qwen3" \
+  pewds@192.168.1.12
+```
+
+**Debug loop pattern:** when a serve is failing, the productive sequence is
+
+1. `cookbook tasks` → find the failing sessionId.
+2. `cookbook output SID 600` → read the last 600 lines, find the actual root-cause line (often above the visible tail because tmux scrollback rolled — request a larger `tail` if the error references "above").
+3. `cookbook stop SID` — kill the previous attempt before relaunching; two serves on the same `--port` collide.
+4. `cookbook serve repo "new cmd"` — try the next variation. Wait ~20s, then `cookbook output` on the new sessionId.
+
+**Hard limits this surface enforces:**
+- `cookbook serve` cmd allowlist + shell-metacharacter rejection — you cannot run arbitrary shell, only model-server binaries.
+- `cookbook stop` only targets task sessionIds matching `[a-zA-Z0-9_-]+`.
+- The agent CAN spawn GPU-pinning long-lived processes — always `cookbook stop` your previous attempt before relaunching, and check `cookbook tasks` for collisions on the same `--port` before launching.
+
 ## Forbidden Bypass Pattern
 
 If you are about to reach the Odysseus host/container, import app internals, query the database, or call MCP helper modules directly, stop. Those paths bypass Odysseus Settings and token scopes. Ask the user to enable the relevant Claude Agent tool toggle instead.

integrations/claude/skills/odysseus/scripts/odysseus_api.py

@@ -17,6 +17,15 @@ def _usage() -> int:
     print("  odysseus_api.py todos add TITLE", file=sys.stderr)
     print("  odysseus_api.py emails list [limit]", file=sys.stderr)
     print("  odysseus_api.py emails read UID", file=sys.stderr)
+    print("  odysseus_api.py cookbook tasks", file=sys.stderr)
+    print("  odysseus_api.py cookbook servers", file=sys.stderr)
+    print("  odysseus_api.py cookbook cached [HOST]", file=sys.stderr)
+    print("  odysseus_api.py cookbook presets", file=sys.stderr)
+    print("  odysseus_api.py cookbook output SESSION_ID [tail]", file=sys.stderr)
+    print("  odysseus_api.py cookbook serve REPO_ID 'CMD' [REMOTE_HOST]", file=sys.stderr)
+    print("  odysseus_api.py cookbook preset NAME", file=sys.stderr)
+    print("  odysseus_api.py cookbook adopt SESSION_ID MODEL [HOST] [PORT]", file=sys.stderr)
+    print("  odysseus_api.py cookbook stop SESSION_ID", file=sys.stderr)
     print("  odysseus_api.py METHOD /api/codex/path [json-body]", file=sys.stderr)
     return 2
 
@@ -72,6 +81,61 @@ def main() -> int:
             body = None
         else:
             return _usage()
+    elif command == "cookbook":
+        if len(sys.argv) < 3:
+            return _usage()
+        action = sys.argv[2].lower()
+        if action == "tasks":
+            method = "GET"
+            path = "/api/codex/cookbook/tasks"
+            body = None
+        elif action == "servers":
+            method = "GET"
+            path = "/api/codex/cookbook/servers"
+            body = None
+        elif action == "output" and len(sys.argv) >= 4:
+            method = "GET"
+            sid = sys.argv[3]
+            tail = sys.argv[4] if len(sys.argv) >= 5 else "400"
+            path = f"/api/codex/cookbook/output/{sid}?tail={tail}"
+            body = None
+        elif action == "cached":
+            method = "GET"
+            if len(sys.argv) >= 4:
+                from urllib.parse import quote
+                path = f"/api/codex/cookbook/cached?host={quote(sys.argv[3])}"
+            else:
+                path = "/api/codex/cookbook/cached"
+            body = None
+        elif action == "presets":
+            method = "GET"
+            path = "/api/codex/cookbook/presets"
+            body = None
+        elif action == "preset" and len(sys.argv) >= 4:
+            from urllib.parse import quote
+            method = "POST"
+            path = f"/api/codex/cookbook/preset/{quote(sys.argv[3])}"
+            body = None
+        elif action == "adopt" and len(sys.argv) >= 5:
+            method = "POST"
+            path = "/api/codex/cookbook/adopt"
+            payload = {"tmux_session": sys.argv[3], "model": sys.argv[4]}
+            if len(sys.argv) >= 6: payload["host"] = sys.argv[5]
+            if len(sys.argv) >= 7: payload["port"] = int(sys.argv[6])
+            body = json.dumps(payload)
+        elif action == "serve" and len(sys.argv) >= 5:
+            method = "POST"
+            path = "/api/codex/cookbook/serve"
+            payload = {"repo_id": sys.argv[3], "cmd": sys.argv[4]}
+            if len(sys.argv) >= 6:
+                payload["remote_host"] = sys.argv[5]
+            body = json.dumps(payload)
+        elif action == "stop" and len(sys.argv) >= 4:
+            method = "POST"
+            path = f"/api/codex/cookbook/stop/{sys.argv[3]}"
+            body = None
+        else:
+            return _usage()
     else:
         if len(sys.argv) < 3:
             return _usage()

integrations/codex/scripts/odysseus_api.py

@@ -17,6 +17,15 @@ def _usage() -> int:
     print("  odysseus_api.py todos add TITLE", file=sys.stderr)
     print("  odysseus_api.py emails list [limit]", file=sys.stderr)
     print("  odysseus_api.py emails read UID", file=sys.stderr)
+    print("  odysseus_api.py cookbook tasks", file=sys.stderr)
+    print("  odysseus_api.py cookbook servers", file=sys.stderr)
+    print("  odysseus_api.py cookbook cached [HOST]", file=sys.stderr)
+    print("  odysseus_api.py cookbook presets", file=sys.stderr)
+    print("  odysseus_api.py cookbook output SESSION_ID [tail]", file=sys.stderr)
+    print("  odysseus_api.py cookbook serve REPO_ID 'CMD' [REMOTE_HOST]", file=sys.stderr)
+    print("  odysseus_api.py cookbook preset NAME", file=sys.stderr)
+    print("  odysseus_api.py cookbook adopt SESSION_ID MODEL [HOST] [PORT]", file=sys.stderr)
+    print("  odysseus_api.py cookbook stop SESSION_ID", file=sys.stderr)
     print("  odysseus_api.py METHOD /api/codex/path [json-body]", file=sys.stderr)
     return 2
 
@@ -72,6 +81,61 @@ def main() -> int:
             body = None
         else:
             return _usage()
+    elif command == "cookbook":
+        if len(sys.argv) < 3:
+            return _usage()
+        action = sys.argv[2].lower()
+        if action == "tasks":
+            method = "GET"
+            path = "/api/codex/cookbook/tasks"
+            body = None
+        elif action == "servers":
+            method = "GET"
+            path = "/api/codex/cookbook/servers"
+            body = None
+        elif action == "output" and len(sys.argv) >= 4:
+            method = "GET"
+            sid = sys.argv[3]
+            tail = sys.argv[4] if len(sys.argv) >= 5 else "400"
+            path = f"/api/codex/cookbook/output/{sid}?tail={tail}"
+            body = None
+        elif action == "cached":
+            method = "GET"
+            if len(sys.argv) >= 4:
+                from urllib.parse import quote
+                path = f"/api/codex/cookbook/cached?host={quote(sys.argv[3])}"
+            else:
+                path = "/api/codex/cookbook/cached"
+            body = None
+        elif action == "presets":
+            method = "GET"
+            path = "/api/codex/cookbook/presets"
+            body = None
+        elif action == "preset" and len(sys.argv) >= 4:
+            from urllib.parse import quote
+            method = "POST"
+            path = f"/api/codex/cookbook/preset/{quote(sys.argv[3])}"
+            body = None
+        elif action == "adopt" and len(sys.argv) >= 5:
+            method = "POST"
+            path = "/api/codex/cookbook/adopt"
+            payload = {"tmux_session": sys.argv[3], "model": sys.argv[4]}
+            if len(sys.argv) >= 6: payload["host"] = sys.argv[5]
+            if len(sys.argv) >= 7: payload["port"] = int(sys.argv[6])
+            body = json.dumps(payload)
+        elif action == "serve" and len(sys.argv) >= 5:
+            method = "POST"
+            path = "/api/codex/cookbook/serve"
+            payload = {"repo_id": sys.argv[3], "cmd": sys.argv[4]}
+            if len(sys.argv) >= 6:
+                payload["remote_host"] = sys.argv[5]
+            body = json.dumps(payload)
+        elif action == "stop" and len(sys.argv) >= 4:
+            method = "POST"
+            path = f"/api/codex/cookbook/stop/{sys.argv[3]}"
+            body = None
+        else:
+            return _usage()
     else:
         if len(sys.argv) < 3:
             return _usage()

integrations/codex/skills/odysseus/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: odysseus
-description: Use when the user asks Codex to read or write Odysseus data from a terminal Codex session through the scoped Codex Agent API. Requires ODYSSEUS_URL and ODYSSEUS_API_TOKEN.
+description: Use when the user asks Codex to read or write Odysseus data (todos, email, calendar, memory, documents) or to launch/monitor/stop a Cookbook model-serve task through the scoped Codex Agent API. Requires ODYSSEUS_URL and ODYSSEUS_API_TOKEN.
 ---
 
 # Odysseus
@@ -105,6 +105,37 @@ python3 integrations/codex/scripts/odysseus_api.py POST /api/codex/memory '{"tex
 - `POST /api/codex/emails/draft` — body matches `SendEmailRequest` (`to`, `cc`, `bcc`, `subject`, `body`, `body_html`, `attachments`, `account_id`, `in_reply_to`, `references`). Requires `email:draft` (or `email:send`).
 - `POST /api/codex/emails/send` — same body. Requires `email:send`. Never send without explicit user instruction.
 
+## Cookbook serve (debug a failing model launch)
+
+The Cookbook surface lets you reproduce what a human would do in Odysseus → Cookbook: read which serves are running, tail their tmux output to see why they crashed, edit the launch command, relaunch, kill a stuck one. Use this when the user is debugging a model server that won't come up (compute-capability errors, OOM, missing kernels, wrong attention backend, etc.).
+
+- `GET /api/codex/cookbook/tasks` — list active serve/download/install tasks (sessionId, type, status, repo_id, remoteHost, payload._cmd). Requires `cookbook:read`.
+- `GET /api/codex/cookbook/servers` — list configured servers (name, host, port, env type + path, model dirs). Requires `cookbook:read`.
+- `GET /api/codex/cookbook/cached?host=<NAME>` — list models already cached on the named server (HF cache + Ollama + extra modelDirs). Call BEFORE `serve` to see what's already on disk. Requires `cookbook:read`.
+- `GET /api/codex/cookbook/presets` — list saved serve presets (model + host + port + cmd). The user's saved preset usually has a working cmd — try `preset NAME` before composing your own. Requires `cookbook:read`.
+- `GET /api/codex/cookbook/output/{session_id}?tail=400` — read the last N lines of the task's persistent log file (preferred) or tmux pane (fallback). The log file persists across vllm crashes, so this returns the actual Python traceback even after the bash prompt + neofetch banner overwrites the pane. Default tail=400. Requires `cookbook:read`.
+- `POST /api/codex/cookbook/serve` — launch a serve task. Body matches `ServeRequest`: `{ repo_id, cmd, remote_host?, ssh_port?, env_prefix?, gpus?, platform? }`. The `cmd` is validated: leading binary must be `vllm`/`python3`/`sglang`/`llama-server`/`ollama`/`node`/`npx`. NEVER prefix with `cd …`, `source …`, or chain with `&&`/`||`/`;`/`$(...)` — the validator rejects shell metacharacters. The venv activation (`env_prefix`) is added automatically from the host's saved settings, so pass the bare binary + args. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/preset/{name}` — launch a saved preset by name. Reuses the working cmd + host the user already saved. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/adopt` — register an externally-launched tmux session into cookbook tracking. Body: `{ tmux_session, model, host?, port? }`. Use this when serve_model rejected a cmd and you fell back to direct ssh+tmux — without adoption, the session is invisible to the UI. Requires `cookbook:launch`.
+- `POST /api/codex/cookbook/stop/{session_id}` — kill the tmux session. Requires `cookbook:launch`.
+
+```bash
+python3 ~/plugins/odysseus/scripts/odysseus_api.py cookbook tasks
+python3 ~/plugins/odysseus/scripts/odysseus_api.py cookbook output serve-abc12345 400
+python3 ~/plugins/odysseus/scripts/odysseus_api.py cookbook stop serve-abc12345
+python3 ~/plugins/odysseus/scripts/odysseus_api.py cookbook serve \
+  /mnt/HADES/models/Qwen3.5-397B-A17B-AWQ \
+  "vllm serve /mnt/HADES/models/Qwen3.5-397B-A17B-AWQ --host 0.0.0.0 --port 8001 --tensor-parallel-size 8 --max-model-len 262144 --gpu-memory-utilization 0.90 --dtype auto --max-num-seqs 8 --trust-remote-code --enable-expert-parallel --enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser qwen3" \
+  pewds@192.168.1.12
+```
+
+**Debug loop pattern:** `tasks` → `output SID 600` (find root cause; request larger `tail` if it references "above") → `stop SID` → `serve repo "new cmd"` → wait ~20s → `output` on the new sessionId.
+
+**Hard limits this surface enforces:**
+- `cookbook serve` cmd allowlist + shell-metacharacter rejection.
+- `cookbook stop` requires sessionIds matching `[a-zA-Z0-9_-]+`.
+- Agent CAN spawn GPU-pinning long-lived processes — always `cookbook stop` your previous attempt before relaunching.
+
 ## Forbidden Bypass Pattern
 
 If you are about to reach the Odysseus host/container, import app internals, query the database, or call MCP helper modules directly, stop. Those paths bypass Odysseus Settings and token scopes. Ask the user to enable the relevant Codex Agent tool toggle instead.

launch-windows.ps1

@@ -30,14 +30,26 @@ function Fail($msg) {
     exit 1
 }
 
+function Test-WindowsBashStub($path) {
+    if (-not $path) { return $false }
+    $lowered = $path.ToLowerInvariant()
+    foreach ($stub in @("system32\bash.exe", "sysnative\bash.exe", "windowsapps\bash.exe")) {
+        if ($lowered.Contains($stub)) { return $true }
+    }
+    return $false
+}
+
 function Find-GitBash {
     $cmd = Get-Command bash -ErrorAction SilentlyContinue
-    if ($cmd) { return $cmd.Source }
+    if ($cmd -and -not (Test-WindowsBashStub $cmd.Source)) { return $cmd.Source }
 
     $roots = @()
     foreach ($name in @("ProgramFiles", "ProgramW6432", "ProgramFiles(x86)", "LocalAppData")) {
         $base = [Environment]::GetEnvironmentVariable($name)
-        if ($base) { $roots += (Join-Path $base "Git") }
+        if ($base) {
+            $roots += (Join-Path $base "Git")
+            if ($name -eq "LocalAppData") { $roots += (Join-Path $base "Programs\Git") }
+        }
     }
     $roots += @("C:\Program Files\Git", "C:\Program Files (x86)\Git")
 

mcp_servers/_common.py

@@ -1,22 +0,0 @@
-"""
-_common.py
-
-Shared constants and helpers for built-in MCP servers.
-"""
-
-MAX_OUTPUT_CHARS = 10_000
-MAX_READ_CHARS = 20_000
-SHELL_TIMEOUT = 60
-PYTHON_TIMEOUT = 30
-SEARCH_TIMEOUT = 30
-
-
-def truncate(text: str, limit: int = MAX_OUTPUT_CHARS) -> str:
-    """Truncate text to *limit* characters with a suffix note."""
-    if not isinstance(text, str):
-        # Tool output is occasionally None or a non-string; len(None) would
-        # raise. Coerce so this shared helper never crashes a tool response.
-        text = "" if text is None else str(text)
-    if len(text) > limit:
-        return text[:limit] + f"\n... (truncated, {len(text)} chars total)"
-    return text

mcp_servers/image_gen_server.py

@@ -16,6 +16,8 @@ from mcp.types import Tool, TextContent
 
 sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
 
+from src.constants import GENERATED_IMAGES_DIR
+
 server = Server("image_gen")
 
 
@@ -115,14 +117,18 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
 
             img = images[0]
             image_url = None
+            # Prefix the instance's public base URL (existing app_public_url setting) so the
+            # link is fully-qualified and clickable when the model echoes it. Empty = relative
+            # same-origin path (unchanged default).
+            _pub_base = (get_setting("app_public_url", "") or "").rstrip("/")
 
             if img.get("b64_json"):
-                img_dir = Path("data/generated_images")
+                img_dir = Path(GENERATED_IMAGES_DIR)
                 img_dir.mkdir(parents=True, exist_ok=True)
                 filename = f"{uuid.uuid4().hex[:12]}.png"
                 img_path = img_dir / filename
                 img_path.write_bytes(base64.b64decode(img["b64_json"]))
-                image_url = f"/api/generated-image/{filename}"
+                image_url = f"{_pub_base}/api/generated-image/{filename}"
 
                 # Save to gallery
                 try:
@@ -146,7 +152,13 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
             else:
                 return [TextContent(type="text", text="Error: Unexpected image API response format")]
 
-            result = f"Generated image for: {prompt[:100]}\nimage_url: {image_url}\nmodel: {model_id}\nsize: {size}"
+            # "Direct link:" rather than an "image_url:" label — small models copied the
+            # label token ("image_url") into the link href, producing a broken link.
+            result = (
+                f"Generated image for: {prompt[:100]}\n"
+                f"Direct link: {image_url}\n"
+                f"model: {model_id}\nsize: {size}"
+            )
             return [TextContent(type="text", text=result)]
 
     except httpx.TimeoutException:

package-lock.json

@@ -1,5 +1,5 @@
 {
-  "name": "odysseus-ui",
+  "name": "odysseus",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {

pyproject.toml

@@ -1,3 +1,22 @@
 [tool.pytest.ini_options]
 testpaths = ["tests"]
 asyncio_mode = "auto"
+# Test-taxonomy markers added at collection time by tests/conftest.py. The
+# stable area_* markers are declared here; the dynamic sub_<filename-token>
+# markers are registered before collection by pytest_configure in
+# tests/conftest.py, so unknown-mark warnings still flag genuine typos outside
+# the taxonomy. See tests/_taxonomy.py and tests/README.md.
+markers = [
+    "area_security: tests covering auth, owner-scope, SSRF, XSS, confinement, redaction",
+    "area_routes: tests covering HTTP route / API behavior",
+    "area_services: tests covering service-layer behavior (llm, cookbook, email, calendar, ...)",
+    "area_cli: tests covering CLI / script behavior",
+    "area_js: JavaScript / Node-backed tests",
+    "area_helpers: self-tests for the shared test helpers in tests/helpers/",
+    "area_unit: pure parser / utility tests that do not clearly belong elsewhere",
+    "area_uncategorized: tests not yet matched by the taxonomy (fallback)",
+    # Fast-lane marker (issue #3443). Opt-in and orthogonal to the area_*/sub_*
+    # taxonomy. The fast lane runs `not slow`; mark a test slow only with
+    # duration evidence (see tests/run_focus.py --durations and tests/README.md).
+    "slow: opt-in marker for known-slow tests; excluded by the fast lane (not slow)",
+]

requirements-optional.txt

@@ -15,7 +15,7 @@ faster-whisper
 # DuckDuckGo as a search provider option.
 # Install if you want DDG in the search-provider dropdown.
 # Alternatives: SearXNG, Brave, Tavily, Serper, Google PSE.
-duckduckgo-search
+ddgs
 
 # PDF form-filling feature (fillable AcroForm detection, field extraction,
 # value/annotation/signature stamping, page rendering for the form overlay).

requirements.txt

@@ -43,3 +43,7 @@ qrcode[pil]
 croniter
 pytest
 pytest-asyncio
+# starlette.testclient prefers httpx2 since Starlette 1.2.0 and warns on every
+# TestClient import when only classic httpx is present. Runtime code keeps
+# using `httpx` above; this is test-client only.
+httpx2

routes/_validators.py

@@ -0,0 +1,31 @@
+import re
+
+from fastapi import HTTPException
+
+
+_REMOTE_HOST_RE = re.compile(
+    r"^(?:[A-Za-z0-9][A-Za-z0-9._-]*@)?[A-Za-z0-9][A-Za-z0-9._-]*$"
+)
+_SSH_PORT_RE = re.compile(r"^\d{1,5}$")
+
+
+def validate_remote_host(v: str | None) -> str | None:
+    if v is None or v == "":
+        return None
+    if not _REMOTE_HOST_RE.match(v):
+        raise HTTPException(
+            400,
+            "Invalid remote_host — must be host or user@host, no SSH option syntax",
+        )
+    return v
+
+
+def validate_ssh_port(v: str | None) -> str | None:
+    if v is None or v == "":
+        return None
+    if not _SSH_PORT_RE.fullmatch(str(v)):
+        raise HTTPException(400, "Invalid ssh_port")
+    port = int(v)
+    if port < 1 or port > 65535:
+        raise HTTPException(400, "Invalid ssh_port")
+    return str(port)

routes/admin_wipe_routes.py

@@ -31,7 +31,7 @@ from core.database import (
     CalendarEvent,
     CalendarCal,
 )
-from src.constants import DATA_DIR
+from src.constants import DATA_DIR, SKILLS_DIR, SKILLS_FILE, GALLERY_DIR, GALLERY_UPLOADS_DIR
 
 logger = logging.getLogger(__name__)
 
@@ -107,7 +107,7 @@ def setup_admin_wipe_routes(session_manager):
                 # Skills live as SKILL.md files under data/skills/. Drop
                 # the entire directory; the SkillsManager re-creates the
                 # tree on next write.
-                skills_dir = os.path.join(DATA_DIR, "skills")
+                skills_dir = SKILLS_DIR
                 count = 0
                 if os.path.isdir(skills_dir):
                     # Count SKILL.md files for the response — quick walk.
@@ -115,7 +115,7 @@ def setup_admin_wipe_routes(session_manager):
                         count += sum(1 for f in files if f == "SKILL.md")
                     _rmtree_quiet(skills_dir)
                 # Legacy fallback file
-                legacy = os.path.join(DATA_DIR, "skills.json")
+                legacy = SKILLS_FILE
                 if os.path.exists(legacy):
                     try:
                         os.remove(legacy)
@@ -151,8 +151,8 @@ def setup_admin_wipe_routes(session_manager):
                 db.query(GalleryAlbum).delete()
                 db.commit()
                 # Also drop the upload dir so disk doesn't keep orphans.
-                _rmtree_quiet(os.path.join(DATA_DIR, "gallery"))
-                _rmtree_quiet(os.path.join(DATA_DIR, "gallery_uploads"))
+                _rmtree_quiet(GALLERY_DIR)
+                _rmtree_quiet(GALLERY_UPLOADS_DIR)
                 return {"status": "deleted", "kind": kind, "count": count}
 
             if kind == "calendar":

routes/api_token_routes.py

@@ -25,6 +25,8 @@ ALLOWED_SCOPES = {
     "calendar:write",
     "memory:read",
     "memory:write",
+    "cookbook:read",
+    "cookbook:launch",
 }
 TOKEN_PROFILES = {
     "chat": ["chat"],
@@ -65,6 +67,7 @@ def _normalize_scopes(scopes: str | list[str] | None = None, profile: str | None
     ensure_before("calendar:write", "calendar:read")
     ensure_before("memory:write", "memory:read")
     ensure_before("email:draft", "email:read")
+    ensure_before("cookbook:launch", "cookbook:read")
 
     return normalized or [DEFAULT_SCOPES]
 
@@ -151,26 +154,37 @@ def setup_api_token_routes() -> APIRouter:
     @router.patch("/tokens/{token_id}")
     async def update_token(request: Request, token_id: str):
         require_admin(request)
+        current_user = get_current_user(request)
         try:
             payload = await request.json()
         except Exception:
             payload = {}
-        scope_list = _normalize_scopes(payload.get("scopes"))
-        scopes_value = ",".join(scope_list)
         with get_db_session() as db:
             token = db.query(ApiToken).filter(ApiToken.id == token_id).first()
             if not token:
                 raise HTTPException(404, "Token not found")
+            if current_user and token.owner != current_user:
+                raise HTTPException(403, "Not your token")
             if isinstance(payload.get("name"), str) and payload["name"].strip():
                 token.name = payload["name"].strip()[:MAX_NAME_LEN]
-            token.scopes = scopes_value
+            # Only touch scopes when the caller actually sent them. A partial
+            # update such as a rename ({"name": ...} with no "scopes" key) must
+            # not silently reset the token to the default scope — that dropped
+            # every previously granted scope.
+            if "scopes" in payload:
+                token.scopes = ",".join(_normalize_scopes(payload.get("scopes")))
             db.add(token)
+            current_scopes = [
+                s.strip()
+                for s in (getattr(token, "scopes", "") or DEFAULT_SCOPES).split(",")
+                if s.strip()
+            ]
             response = {
                 "id": token_id,
                 "name": getattr(token, "name", ""),
                 "owner": getattr(token, "owner", None),
                 "token_prefix": getattr(token, "token_prefix", ""),
-                "scopes": scope_list,
+                "scopes": current_scopes,
             }
         _invalidate_cache(request)
         return response
@@ -178,10 +192,14 @@ def setup_api_token_routes() -> APIRouter:
     @router.delete("/tokens/{token_id}")
     def delete_token(request: Request, token_id: str):
         require_admin(request)
+        current_user = get_current_user(request)
         with get_db_session() as db:
-            deleted = db.query(ApiToken).filter(ApiToken.id == token_id).delete()
-            if not deleted:
+            token = db.query(ApiToken).filter(ApiToken.id == token_id).first()
+            if not token:
                 raise HTTPException(404, "Token not found")
+            if current_user and token.owner != current_user:
+                raise HTTPException(403, "Not your token")
+            db.delete(token)
         _invalidate_cache(request)
         return {"status": "deleted"}
 

routes/auth_routes.py

@@ -7,7 +7,13 @@ import asyncio
 import logging
 import os
 
+import json
+import re
+from pathlib import Path
+
+from core.atomic_io import atomic_write_json, atomic_write_text
 from core.auth import AuthManager
+from src.constants import DEEP_RESEARCH_DIR, MEMORY_FILE, SKILLS_DIR
 from src.rate_limiter import RateLimiter
 from src.settings_scrub import scrub_settings
 from src.settings import (
@@ -131,10 +137,8 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
                 return {"ok": False, "requires_totp": True, "username": username}
             if not auth_manager.totp_verify(username, body.totp_code):
                 raise HTTPException(401, "Invalid 2FA code")
-        # All checks passed — create session
-        token = await asyncio.to_thread(auth_manager.create_session, username, body.password)
-        if not token:
-            raise HTTPException(401, "Invalid credentials")
+        # All checks passed — create session (password already verified above)
+        token = await asyncio.to_thread(auth_manager.create_session_trusted, username)
         cookie_kwargs = dict(
             key=SESSION_COOKIE,
             value=token,
@@ -293,9 +297,30 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         if new_username in auth_manager.users:
             raise HTTPException(409, "Username already taken")
 
+        # Gate on auth first. Every mutation below is contingent on this
+        # succeeding — doing it last meant a rejected rename (e.g. reserved
+        # username) left file-backed owner fields already rewritten with no
+        # way to roll them back.
+        ok = auth_manager.rename_user(old_username, new_username, user)
+        if not ok:
+            raise HTTPException(400, "Cannot rename user")
+
+        def _rollback_auth_rename() -> bool:
+            # On self-rename the admin session has already moved to the new
+            # username, so the rollback must authenticate as the new user.
+            rollback_user = new_username if user == old_username else user
+            try:
+                return bool(auth_manager.rename_user(new_username, old_username, rollback_user))
+            except Exception as rollback_err:
+                logger.error(
+                    "Failed to roll back auth rename %s -> %s after owner migration failure: %s",
+                    new_username, old_username, rollback_err,
+                )
+                return False
+
         # Usernames are ownership keys for user data. Rename the common
-        # owner-scoped DB rows before changing auth so the account keeps
-        # access to its sessions, docs, email accounts, tasks, etc.
+        # owner-scoped DB rows so the account keeps access to its sessions,
+        # docs, email accounts, tasks, etc.
         try:
             from sqlalchemy import func
             from core.database import Base, SessionLocal
@@ -318,6 +343,11 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
                 db.close()
         except Exception as e:
             logger.error("Failed to rename owner references %s -> %s: %s", old_username, new_username, e)
+            if not _rollback_auth_rename():
+                logger.error(
+                    "Auth rename %s -> %s could not be rolled back after owner migration failure",
+                    old_username, new_username,
+                )
             raise HTTPException(500, "Failed to rename user data")
 
         # Per-user prefs are JSON-backed, not SQL-backed.
@@ -337,9 +367,116 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         except Exception as e:
             logger.warning("Failed to rename user prefs %s -> %s: %s", old_username, new_username, e)
 
-        ok = auth_manager.rename_user(old_username, new_username, user)
-        if not ok:
-            raise HTTPException(400, "Cannot rename user")
+        # In-flight deep-research tasks live in the process-local
+        # ResearchHandler registry. They are not covered by the persisted JSON
+        # migration above, but the research routes filter and cancel by this
+        # owner field while the job is running. Do this before sweeping
+        # completed JSON files so a job that finishes during the rename saves
+        # with the new owner or is caught by the disk sweep below.
+        try:
+            rh = getattr(request.app.state, "research_handler", None)
+            rename_owner = getattr(rh, "rename_owner", None)
+            if callable(rename_owner):
+                rename_owner(old_username, new_username)
+        except Exception as e:
+            logger.warning("Failed to rename active research tasks %s -> %s: %s", old_username, new_username, e)
+
+        # deep_research: each completed report is a standalone JSON file with
+        # an `owner` field. research_routes filters by d.get("owner") == user,
+        # so a stale owner makes every report invisible to the renamed user.
+        try:
+            dr_dir = Path(DEEP_RESEARCH_DIR)
+            if dr_dir.is_dir():
+                for p in dr_dir.glob("*.json"):
+                    try:
+                        d = json.loads(p.read_text(encoding="utf-8"))
+                        if str(d.get("owner", "")).strip().lower() == old_username:
+                            d["owner"] = new_username
+                            atomic_write_json(str(p), d)
+                    except Exception as err:
+                        logger.warning("Failed to update research owner in %s: %s", p.name, err)
+        except Exception as e:
+            logger.warning("Failed to rename research owner references %s -> %s: %s", old_username, new_username, e)
+
+        # memory.json: a flat JSON array where each entry carries an `owner`
+        # field. memory_manager.load(owner=user) filters on it, so stale
+        # entries disappear from the memory panel.
+        try:
+            if os.path.isfile(MEMORY_FILE):
+                with open(MEMORY_FILE, encoding="utf-8") as fh:
+                    entries = json.loads(fh.read())
+                if isinstance(entries, list):
+                    changed = False
+                    for entry in entries:
+                        if isinstance(entry, dict) and str(entry.get("owner", "")).strip().lower() == old_username:
+                            entry["owner"] = new_username
+                            changed = True
+                    if changed:
+                        atomic_write_json(MEMORY_FILE, entries)
+        except Exception as e:
+            logger.warning("Failed to rename memory.json owner references %s -> %s: %s", old_username, new_username, e)
+
+        # uploads.json: upload rows use owner metadata for access checks and
+        # owner-prefixed index keys for dedupe. Rename both so attachments keep
+        # resolving after the account username changes.
+        try:
+            upload_handler = getattr(request.app.state, "upload_handler", None)
+            rename_owner = getattr(upload_handler, "rename_owner", None)
+            if callable(rename_owner):
+                rename_owner(old_username, new_username)
+        except Exception as e:
+            logger.warning("Failed to rename upload owner references %s -> %s: %s", old_username, new_username, e)
+
+        # skills: SKILL.md frontmatter carries owner: <username>; the usage
+        # sidecar (_usage.json) keys entries as owner::skill-name. Both must
+        # be updated or the renamed user's Skills panel goes empty.
+        try:
+            skills_root = Path(SKILLS_DIR)
+            if skills_root.is_dir():
+                _owner_re = re.compile(
+                    r'(?m)^(owner:\s*)' + re.escape(old_username) + r'\s*$',
+                    re.IGNORECASE,
+                )
+                for p in skills_root.rglob("SKILL.md"):
+                    try:
+                        text = p.read_text(encoding="utf-8")
+                        new_text = _owner_re.sub(r'\g<1>' + new_username, text)
+                        if new_text != text:
+                            atomic_write_text(str(p), new_text)
+                    except Exception as err:
+                        logger.warning("Failed to update skill owner in %s: %s", p, err)
+                usage_path = skills_root / "_usage.json"
+                if usage_path.is_file():
+                    try:
+                        usage = json.loads(usage_path.read_text(encoding="utf-8"))
+                        if isinstance(usage, dict):
+                            new_usage = {}
+                            changed = False
+                            for k, v in usage.items():
+                                owner_part, sep, skill_part = k.partition("::")
+                                if sep and owner_part.lower() == old_username:
+                                    new_usage[new_username + "::" + skill_part] = v
+                                    changed = True
+                                else:
+                                    new_usage[k] = v
+                            if changed:
+                                atomic_write_json(str(usage_path), new_usage)
+                    except Exception as err:
+                        logger.warning("Failed to update skills usage keys %s -> %s: %s", old_username, new_username, err)
+        except Exception as e:
+            logger.warning("Failed to rename skills owner references %s -> %s: %s", old_username, new_username, e)
+
+        # The in-memory session cache (session_manager.sessions) stores each
+        # session's owner at load time. Without this patch the renamed user's
+        # sessions are invisible on the next /api/sessions call because
+        # get_sessions_for_user does an exact `s.owner == username` comparison
+        # against stale in-memory values.
+        sm = getattr(request.app.state, "session_manager", None)
+        if sm is not None:
+            for sess in list(getattr(sm, "sessions", {}).values()):
+                if str(getattr(sess, "owner", None) or "").strip().lower() == old_username:
+                    sess.owner = new_username
+
         # The owner-rename loop above updated ApiToken.owner in the DB, but the
         # bearer-token cache still maps each token to the OLD owner. Without
         # refreshing it, the renamed user's API tokens resolve to the old (now
@@ -380,7 +517,23 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         user = _get_current_user(request)
         if not user or not auth_manager.is_admin(user):
             raise HTTPException(403, "Admin only")
-        ok = auth_manager.delete_user(body.username, user)
+
+        def _invalidate_api_token_cache():
+            try:
+                invalidator = getattr(request.app.state, "invalidate_token_cache", None)
+                if invalidator:
+                    invalidator()
+            except Exception:
+                pass
+
+        try:
+            ok = auth_manager.delete_user(body.username, user)
+        except Exception:
+            # delete_user can touch ApiToken rows before a later auth-store write
+            # fails. Dirty the bearer cache anyway so a partial token purge does
+            # not leave already-cached tokens authenticating until restart.
+            _invalidate_api_token_cache()
+            raise
         if not ok:
             raise HTTPException(400, "Cannot delete user")
         # delete_user removes the user's ApiToken rows, but the bearer-auth
@@ -388,12 +541,7 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
         # rebuilds when flagged dirty. Without this, a deleted user's already
         # cached token keeps authenticating until some other token op or a
         # restart clears the cache. Mirror what the token routes do.
-        try:
-            invalidator = getattr(request.app.state, "invalidate_token_cache", None)
-            if invalidator:
-                invalidator()
-        except Exception:
-            pass
+        _invalidate_api_token_cache()
         return {"ok": True}
 
     # ---- Feature visibility (admin-managed) ----
@@ -585,6 +733,27 @@ def setup_auth_routes(auth_manager: AuthManager) -> APIRouter:
                     hint = " If this is Docker Compose ntfy, set NTFY_BIND to that host/Tailscale IP and NTFY_BASE_URL to the same server URL in .env, then recreate ntfy."
                 return {"ok": False, "message": f"ntfy publish to {full_url} failed: {e}.{hint}"[:500]}
 
+        if preset == "discord_webhook":
+            import httpx
+            webhook_url = (integ.get("base_url") or "").strip()
+            if not webhook_url:
+                return {"ok": False, "message": "No webhook URL set — paste the full Discord webhook URL into the Base URL field."}
+            payload = {
+                "embeds": [{
+                    "title": "Odysseus connectivity test",
+                    "description": "If you see this, your Discord Webhook integration is wired up correctly.",
+                    "color": 5793266,
+                }]
+            }
+            try:
+                async with httpx.AsyncClient(timeout=8.0) as client:
+                    r = await client.post(webhook_url, json=payload)
+                if r.is_success:
+                    return {"ok": True, "message": "Test embed sent — check your Discord channel to confirm it arrived."}
+                return {"ok": False, "message": f"Discord returned HTTP {r.status_code}: {r.text[:200]}"}
+            except Exception as e:
+                return {"ok": False, "message": f"Request failed: {e}"[:400]}
+
         # All other presets: GET against a known health endpoint.
         # Fall back to detecting from name if preset is missing.
         health_paths = {

routes/backup_routes.py

@@ -101,24 +101,74 @@ def setup_backup_routes(memory_manager, preset_manager, skills_manager) -> APIRo
         # ── Skills ──
         if "skills" in body and isinstance(body["skills"], list):
             existing = skills_manager.load_all()
-            existing_ids = {s.get("id") for s in existing}
-            existing_titles = {s.get("title", "").strip().lower() for s in existing}
+            # Dedup against THIS user's own skills only. Using every tenant's
+            # rows (load_all) meant a skill whose id/name/title matched any
+            # other user's was silently skipped, so the importing user lost
+            # their own data — same cross-tenant bug fixed for memories above.
+            # The full store is still saved back below.
+            own = [s for s in existing if s.get("owner") == user]
+            existing_names = {s.get("name") for s in own if s.get("name")}
+            existing_ids = {s.get("id") for s in own if s.get("id")}
+            existing_titles = {
+                (s.get("title") or s.get("description") or "").strip().lower()
+                for s in own
+            }
             added = 0
             for skill in body["skills"]:
-                if not isinstance(skill, dict) or not skill.get("title"):
+                if not isinstance(skill, dict):
                     continue
-                # Skip if same id or same title already exists
-                if skill.get("id") in existing_ids:
+                title = (
+                    skill.get("title") or skill.get("description")
+                    or skill.get("name") or ""
+                ).strip()
+                if not title:
                     continue
-                if skill["title"].strip().lower() in existing_titles:
+                sid = skill.get("id") or skill.get("name")
+                if sid and sid in existing_ids:
                     continue
-                if user and not skill.get("owner"):
-                    skill["owner"] = user
-                existing.append(skill)
-                existing_ids.add(skill.get("id"))
-                existing_titles.add(skill["title"].strip().lower())
+                nm = skill.get("name")
+                if nm and nm in existing_names:
+                    continue
+                if title.lower() in existing_titles:
+                    continue
+                owner = skill.get("owner")
+                if user and not owner:
+                    owner = user
+                # Skills live on disk as SKILL.md files; the old JSON-era
+                # skills_manager.save() no longer exists. Write each new skill
+                # via add_skill (source="user" skips auto-dedup — this is an
+                # explicit backup restore).
+                result = skills_manager.add_skill(
+                    title=title,
+                    name=skill.get("name"),
+                    description=skill.get("description"),
+                    problem=skill.get("problem", ""),
+                    solution=skill.get("solution", ""),
+                    steps=skill.get("steps"),
+                    tags=skill.get("tags"),
+                    source="user",
+                    teacher_model=skill.get("teacher_model"),
+                    confidence=skill.get("confidence", 0.8),
+                    owner=owner,
+                    category=skill.get("category", "general"),
+                    when_to_use=skill.get("when_to_use"),
+                    procedure=skill.get("procedure"),
+                    pitfalls=skill.get("pitfalls"),
+                    verification=skill.get("verification"),
+                    platforms=skill.get("platforms"),
+                    requires_toolsets=skill.get("requires_toolsets"),
+                    fallback_for_toolsets=skill.get("fallback_for_toolsets"),
+                    status=skill.get("status", "draft"),
+                    version=skill.get("version", "1.0.0"),
+                )
+                if result.get("_deduped"):
+                    continue
+                if result.get("name"):
+                    existing_names.add(result["name"])
+                if result.get("id"):
+                    existing_ids.add(result["id"])
+                existing_titles.add(title.lower())
                 added += 1
-            skills_manager.save(existing)
             imported.append(f"{added} skills")
 
         # ── Presets ──

routes/calendar_routes.py

@@ -1,6 +1,7 @@
 """Calendar routes — local SQLite-backed calendar CRUD."""
 
 import logging
+import re
 import uuid
 from datetime import datetime, date, timedelta
 from typing import Optional, List
@@ -12,7 +13,7 @@ from dateutil.rrule import rrulestr
 
 from core.database import SessionLocal, CalendarCal, CalendarEvent
 from src.auth_helpers import require_user
-from src.upload_limits import read_upload_limited
+from src.upload_limits import read_upload_limited, ICS_MAX_BYTES
 
 logger = logging.getLogger(__name__)
 
@@ -100,6 +101,15 @@ def _ics_escape(text: str) -> str:
     )
 
 
+def _safe_ics_filename(name: str) -> str:
+    """Return a conservative .ics filename safe for Content-Disposition."""
+    stem = name if isinstance(name, str) else ""
+    stem = re.sub(r"[^A-Za-z0-9._-]", "_", stem).strip("._-")
+    if not stem:
+        stem = "calendar"
+    return f"{stem[:128]}.ics"
+
+
 def _resolve_base_uid(uid: str) -> str:
     """Extract the base series UID from a compound occurrence UID.
 
@@ -248,6 +258,17 @@ def parse_due_for_user(s: str) -> str:
         if t is not None:
             return base.replace(hour=t[0], minute=t[1]).isoformat()
 
+    # Time-first: "3pm today", "11pm today", "9am tomorrow"
+    m = _re.match(r'^(.+?)\s+(today|tonight|tomorrow|tmrw|yesterday)$', lower)
+    if m:
+        time_part, word = m.group(1).strip(), m.group(2)
+        base = today
+        if word in ("tomorrow", "tmrw"): base = today + _td(days=1)
+        elif word == "yesterday":        base = today - _td(days=1)
+        t = _parse_time(time_part)
+        if t is not None:
+            return base.replace(hour=t[0], minute=t[1]).isoformat()
+
     m = _re.match(r'^in\s+(\d+)\s*(hour|hr|minute|min|day)s?\s*$', lower)
     if m:
         n = int(m.group(1)); unit = m.group(2)
@@ -399,7 +420,17 @@ def _parse_dt(s: str) -> datetime:
     # Last resort: dateutil's fuzzy parser
     try:
         from dateutil import parser as _du
-        return _du.parse(s)
+        parsed = _du.parse(s)
+        # Strip tz like every other return path above — this function's
+        # contract is naive datetimes (CalendarEvent.dtstart is naive). An
+        # offset-bearing non-ISO input (e.g. RFC-2822 "Mon, 05 Jan 2026
+        # 14:00:00 +0900") otherwise leaked tz-aware into the naive column and
+        # crashed read-back comparisons in _expand_rrule with "can't compare
+        # offset-naive and offset-aware datetimes".
+        if parsed.tzinfo is not None:
+            from datetime import timezone as _tz
+            return parsed.astimezone(_tz.utc).replace(tzinfo=None)
+        return parsed
     except Exception:
         raise ValueError(f"could not parse datetime: {s!r}")
 
@@ -440,6 +471,9 @@ def _event_to_dict(ev: CalendarEvent) -> dict:
 
 # ── Recurrence expansion ──
 
+_RRULE_EXPANSION_LIMIT = 1000
+
+
 def _expand_rrule(
     ev: CalendarEvent, start: datetime, end: datetime
 ) -> List[dict]:
@@ -462,6 +496,7 @@ def _expand_rrule(
         d = _event_to_dict(ev)
         d["is_recurrence"] = False
         d["series_uid"] = ev.uid
+        d["truncated"] = False
         return [d]
 
     # Parse the rrule, applying it to the base dtstart.
@@ -487,6 +522,7 @@ def _expand_rrule(
         d = _event_to_dict(ev)
         d["is_recurrence"] = False
         d["series_uid"] = ev.uid
+        d["truncated"] = False
         # Malformed RRULE rows are fetched by the recurring SQL branch
         # with only dtstart < end_dt — the base event may not actually
         # overlap the window. Only return if it does.
@@ -499,22 +535,26 @@ def _expand_rrule(
     # (matching non-recurring overlap semantics: dtstart < end AND
     # dtend > start).
     expand_start = start - duration
-    occurrences = rule.between(expand_start, end, inc=True)
-    if not occurrences:
-        return []
-
     results = []
+    truncated = False
     base = _event_to_dict(ev)
 
-    for occ_start in occurrences:
+    for occ_start in rule.xafter(expand_start, inc=True):
+        if occ_start >= end:
+            break
+
         occ_end = occ_start + duration
 
         # Overlap filter: occurrence must intersect [start, end).
         # This enforces exclusive-end semantics (occ_start >= end is
         # excluded) and includes multi-day crossings (occ_end > start).
-        if occ_start >= end or occ_end <= start:
+        if occ_end <= start:
             continue
 
+        if len(results) >= _RRULE_EXPANSION_LIMIT:
+            truncated = True
+            break
+
         # Build the compound uid: {base_uid}::{date} or ::{datetime}
         if ev.all_day:
             occ_uid = f"{ev.uid}::{occ_start.strftime('%Y-%m-%d')}"
@@ -525,6 +565,7 @@ def _expand_rrule(
         d["uid"] = occ_uid
         d["series_uid"] = ev.uid
         d["is_recurrence"] = True
+        d["truncated"] = False
 
         if ev.all_day:
             d["dtstart"] = occ_start.strftime("%Y-%m-%d")
@@ -537,6 +578,10 @@ def _expand_rrule(
 
         results.append(d)
 
+    if truncated:
+        for d in results:
+            d["truncated"] = True
+
     return results
 
 
@@ -545,72 +590,178 @@ def _expand_rrule(
 def setup_calendar_routes() -> APIRouter:
     router = APIRouter(prefix="/api/calendar", tags=["calendar"])
 
-    # CalDAV connect form (Integrations → Calendar). Storage is local
-    # SQLite; sync (src/caldav_sync.py) pulls remote events into it on
-    # calendar open and periodically via the scheduler.
+    # ── CalDAV multi-account helpers ─────────────────────────────────────────
+
+    def _get_caldav_accounts(owner: str) -> list:
+        from src.caldav_sync import _load_caldav_accounts
+        return _load_caldav_accounts(owner)
+
+    def _save_caldav_accounts(owner: str, accounts: list) -> None:
+        from routes.prefs_routes import _load_for_user, _save_for_user
+        prefs = _load_for_user(owner) or {}
+        prefs["caldav_accounts"] = accounts
+        prefs.pop("caldav", None)
+        _save_for_user(owner, prefs)
+
+    # ── CalDAV config routes (backward-compat single-account API) ────────────
+
     @router.get("/config")
     async def get_config(request: Request):
+        """Legacy single-account endpoint — returns the first configured account."""
         owner = _require_user(request)
-        from routes.prefs_routes import _load_for_user
-        cfg = (_load_for_user(owner) or {}).get("caldav", {}) or {}
-        caldav_password = cfg.get("password") or ""
-        if caldav_password:
+        accounts = _get_caldav_accounts(owner)
+        if not accounts:
+            return {"url": "", "username": "", "password": "", "has_password": False, "local": True}
+        first = accounts[0]
+        pw = first.get("password") or ""
+        has_pw = False
+        if pw:
             try:
                 from src.secret_storage import decrypt
-                caldav_password = decrypt(caldav_password)
+                has_pw = bool(decrypt(pw))
             except Exception:
-                pass
-        # Surface url+username but never hand the password back to the
-        # client — saved-state UI shouldn't leak the credential.
+                has_pw = bool(pw)
         return {
-            "url": cfg.get("url", "") or "",
-            "username": cfg.get("username", "") or "",
+            "url": first.get("url", "") or "",
+            "username": first.get("username", "") or "",
             "password": "",
-            "has_password": bool(caldav_password),
-            "local": not bool(cfg.get("url")),
+            "has_password": has_pw,
+            "local": not bool(first.get("url")),
         }
 
     @router.post("/config")
     async def save_config(request: Request):
+        """Legacy single-account endpoint — upserts the first account."""
         owner = _require_user(request)
-        from routes.prefs_routes import _load_for_user, _save_for_user
         try:
             body = await request.json()
         except Exception:
             body = {}
-        prefs = _load_for_user(owner) or {}
-        cfg = dict(prefs.get("caldav") or {})
-        # Empty url => clear the whole entry (treat as "remove integration").
+        accounts = _get_caldav_accounts(owner)
         if not (body.get("url") or "").strip():
-            prefs.pop("caldav", None)
-            _save_for_user(owner, prefs)
+            _save_caldav_accounts(owner, [])
             return {"ok": True, "cleared": True}
         from src.caldav_sync import validate_caldav_url
         try:
-            cfg["url"] = validate_caldav_url(body.get("url", ""))
+            validated_url = validate_caldav_url(body.get("url", ""))
         except ValueError as e:
             raise HTTPException(400, str(e))
-        cfg["username"] = (body.get("username") or "").strip()
-        # Preserve the stored password when the client sends an empty
-        # one (edit form re-submitted without re-typing the password).
-        # cfg already holds the existing (already-encrypted) password from
-        # prefs, so we only touch it when a new password is supplied —
-        # re-encrypting the stored value would double-encrypt it.
+        if accounts:
+            acc = dict(accounts[0])
+        else:
+            import uuid as _uuid
+            acc = {"id": str(_uuid.uuid4()), "label": "CalDAV"}
+        acc["url"] = validated_url
+        acc["username"] = (body.get("username") or "").strip()
         if body.get("password"):
             from src.secret_storage import encrypt
-            cfg["password"] = encrypt(body["password"])
-        prefs["caldav"] = cfg
-        _save_for_user(owner, prefs)
+            acc["password"] = encrypt(body["password"])
+        new_accounts = [acc] + (accounts[1:] if len(accounts) > 1 else [])
+        _save_caldav_accounts(owner, new_accounts)
+        return {"ok": True}
+
+    # ── CalDAV multi-account CRUD ─────────────────────────────────────────────
+
+    @router.get("/config/accounts")
+    async def list_caldav_accounts(request: Request):
+        """Return all configured CalDAV accounts (passwords never returned)."""
+        owner = _require_user(request)
+        accounts = _get_caldav_accounts(owner)
+        safe = []
+        for acc in accounts:
+            pw = acc.get("password") or ""
+            has_pw = False
+            if pw:
+                try:
+                    from src.secret_storage import decrypt
+                    has_pw = bool(decrypt(pw))
+                except Exception:
+                    has_pw = bool(pw)
+            safe.append({
+                "id": acc.get("id", ""),
+                "label": acc.get("label", "") or acc.get("url", ""),
+                "url": acc.get("url", "") or "",
+                "username": acc.get("username", "") or "",
+                "has_password": has_pw,
+            })
+        return {"accounts": safe}
+
+    @router.post("/config/accounts")
+    async def add_caldav_account(request: Request):
+        """Add a new CalDAV account."""
+        import uuid as _uuid
+        owner = _require_user(request)
+        try:
+            body = await request.json()
+        except Exception:
+            body = {}
+        from src.caldav_sync import validate_caldav_url
+        try:
+            url = validate_caldav_url(body.get("url", ""))
+        except ValueError as e:
+            raise HTTPException(400, str(e))
+        if not body.get("password"):
+            raise HTTPException(400, "Password is required")
+        from src.secret_storage import encrypt
+        new_acc = {
+            "id": str(_uuid.uuid4()),
+            "label": (body.get("label") or "").strip() or "CalDAV",
+            "url": url,
+            "username": (body.get("username") or "").strip(),
+            "password": encrypt(body["password"]),
+        }
+        accounts = _get_caldav_accounts(owner)
+        accounts.append(new_acc)
+        _save_caldav_accounts(owner, accounts)
+        return {"ok": True, "id": new_acc["id"]}
+
+    @router.put("/config/accounts/{account_id}")
+    async def update_caldav_account(account_id: str, request: Request):
+        """Update an existing CalDAV account by id."""
+        owner = _require_user(request)
+        try:
+            body = await request.json()
+        except Exception:
+            body = {}
+        accounts = _get_caldav_accounts(owner)
+        idx = next((i for i, a in enumerate(accounts) if a.get("id") == account_id), None)
+        if idx is None:
+            raise HTTPException(404, "Account not found")
+        acc = dict(accounts[idx])
+        if body.get("url"):
+            from src.caldav_sync import validate_caldav_url
+            try:
+                acc["url"] = validate_caldav_url(body["url"])
+            except ValueError as e:
+                raise HTTPException(400, str(e))
+        if body.get("label") is not None:
+            acc["label"] = (body.get("label") or "").strip() or "CalDAV"
+        if body.get("username") is not None:
+            acc["username"] = (body.get("username") or "").strip()
+        if body.get("password"):
+            from src.secret_storage import encrypt
+            acc["password"] = encrypt(body["password"])
+        accounts[idx] = acc
+        _save_caldav_accounts(owner, accounts)
+        return {"ok": True}
+
+    @router.delete("/config/accounts/{account_id}")
+    async def delete_caldav_account(account_id: str, request: Request):
+        """Remove a CalDAV account by id."""
+        owner = _require_user(request)
+        accounts = _get_caldav_accounts(owner)
+        new_accounts = [a for a in accounts if a.get("id") != account_id]
+        if len(new_accounts) == len(accounts):
+            raise HTTPException(404, "Account not found")
+        _save_caldav_accounts(owner, new_accounts)
         return {"ok": True}
 
     @router.post("/test")
     async def test_connection(request: Request):
-        """Actually probe the configured CalDAV server with a PROPFIND
-        request (the same handshake every CalDAV client uses). Accepts
-        an optional {url, username, password} body so the user can test
-        a configuration BEFORE saving it; falls back to the stored
-        creds otherwise. Returns {ok, error?} with a useful message on
-        failure (status code, auth issue, network error)."""
+        """Probe a CalDAV server with a PROPFIND. Accepts an optional body:
+        {url, username, password} to test before saving, or {account_id} to
+        test an already-saved account. Falls back to the first saved account
+        when nothing is provided."""
         owner = _require_user(request)
         try:
             body = await request.json()
@@ -620,19 +771,24 @@ def setup_calendar_routes() -> APIRouter:
         user = (body.get("username") or "").strip()
         pw = body.get("password") or ""
         if not (url and user and pw):
-            # Fall back to saved settings for this user.
-            from routes.prefs_routes import _load_for_user
-            cfg = (_load_for_user(owner) or {}).get("caldav", {}) or {}
-            url = url or (cfg.get("url") or "")
-            user = user or (cfg.get("username") or "")
-            if not pw:
-                pw = cfg.get("password") or ""
-                if pw:
-                    try:
-                        from src.secret_storage import decrypt
-                        pw = decrypt(pw)
-                    except Exception:
-                        pass
+            # Look up a saved account: by id if supplied, else first account.
+            accounts = _get_caldav_accounts(owner)
+            acc = None
+            if body.get("account_id"):
+                acc = next((a for a in accounts if a.get("id") == body["account_id"]), None)
+            if acc is None and accounts:
+                acc = accounts[0]
+            if acc:
+                url = url or (acc.get("url") or "")
+                user = user or (acc.get("username") or "")
+                if not pw:
+                    pw = acc.get("password") or ""
+                    if pw:
+                        try:
+                            from src.secret_storage import decrypt
+                            pw = decrypt(pw)
+                        except Exception:
+                            pass
         if not (url and user and pw):
             return {"ok": False, "error": "Missing URL, username, or password"}
         from src.caldav_sync import validate_caldav_url
@@ -695,6 +851,27 @@ def setup_calendar_routes() -> APIRouter:
         from src.caldav_sync import sync_caldav
         return await sync_caldav(owner)
 
+
+    @router.delete("/calendars/{cal_id}")
+    async def delete_calendar(request: Request, cal_id: str):
+        owner = _require_user(request)
+        db = SessionLocal()
+        try:
+            cal = _get_or_404_calendar(db, cal_id, owner)
+            db.query(CalendarEvent).filter(CalendarEvent.calendar_id == cal_id).delete()
+            db.delete(cal)
+            db.commit()
+            return {"ok": True}
+        except HTTPException:
+            raise
+        except Exception as e:
+            db.rollback()
+            logger.error("Failed to delete calendar %s: %s", cal_id, e)
+            raise HTTPException(500, "Failed to delete calendar")
+        finally:
+            db.close()
+
+
     @router.get("/calendars")
     async def list_calendars(request: Request):
         owner = _require_user(request)
@@ -703,7 +880,7 @@ def setup_calendar_routes() -> APIRouter:
             _ensure_default_calendar(db, owner)
             cals = db.query(CalendarCal).filter(CalendarCal.owner == owner).all()
             return {"calendars": [
-                {"name": c.name, "href": c.id, "color": c.color}
+                {"name": c.name, "href": c.id, "color": c.color, "source": c.source}
                 for c in cals
             ]}
         except HTTPException:
@@ -766,8 +943,12 @@ def setup_calendar_routes() -> APIRouter:
                 expanded.extend(_expand_rrule(e, start_dt, end_dt))
 
             # Sort by occurrence start time for consistent frontend ordering.
+            truncated = any(e.get("truncated") for e in expanded)
             expanded.sort(key=lambda d: d["dtstart"])
-            return {"events": expanded}
+            response: dict = {"events": expanded}
+            if truncated:
+                response["truncated"] = True
+            return response
         except HTTPException:
             raise
         except Exception as e:
@@ -970,27 +1151,10 @@ def setup_calendar_routes() -> APIRouter:
         finally:
             db.close()
 
-    @router.delete("/calendars/{cal_id}")
-    async def delete_calendar(request: Request, cal_id: str):
-        owner = _require_user(request)
-        db = SessionLocal()
-        try:
-            cal = _get_or_404_calendar(db, cal_id, owner)
-            db.query(CalendarEvent).filter(CalendarEvent.calendar_id == cal_id).delete()
-            db.delete(cal)
-            db.commit()
-            return {"ok": True}
-        except HTTPException:
-            raise
-        except Exception as e:
-            db.rollback()
-            return {"error": str(e)}
-        finally:
-            db.close()
 
-    # 10 MB hard cap on ICS upload. Loading the whole file into memory is
-    # unavoidable with python-icalendar, so an unbounded upload would OOM.
-    _ICS_MAX_BYTES = 10 * 1024 * 1024
+    # Hard cap on ICS upload (ICS_MAX_BYTES, default 10 MB). Loading the whole
+    # file into memory is unavoidable with python-icalendar, so an unbounded
+    # upload would OOM.
 
     @router.post("/import")
     async def import_ics(request: Request, file: UploadFile = File(...), calendar_name: str = ""):
@@ -1000,7 +1164,7 @@ def setup_calendar_routes() -> APIRouter:
         owner = _require_user(request)
         db = SessionLocal()
         try:
-            content = await read_upload_limited(file, _ICS_MAX_BYTES, "ICS file")
+            content = await read_upload_limited(file, ICS_MAX_BYTES, "ICS file")
             try:
                 cal_data = iCal.from_ical(content)
             except Exception as e:
@@ -1168,11 +1332,14 @@ def setup_calendar_routes() -> APIRouter:
             lines.append("END:VCALENDAR")
 
             ics_data = "\r\n".join(lines)
-            safe_name = cal.name.replace(" ", "_").replace("/", "_")
+            download_name = _safe_ics_filename(cal.name)
             return Response(
                 content=ics_data,
                 media_type="text/calendar",
-                headers={"Content-Disposition": f'attachment; filename="{safe_name}.ics"'},
+                headers={
+                    "Content-Disposition": f'attachment; filename="{download_name}"',
+                    "X-Content-Type-Options": "nosniff",
+                },
             )
         except HTTPException:
             raise
@@ -1194,7 +1361,7 @@ def setup_calendar_routes() -> APIRouter:
         "tomorrow", "next Tuesday", "in 30 minutes" resolve correctly.
         Uses the "utility" endpoint (small / fast model) to keep latency low.
         """
-        _require_user(request)
+        owner = _require_user(request)
         from src.endpoint_resolver import resolve_endpoint
         from src.llm_core import llm_call_async
         from src.text_helpers import strip_think
@@ -1220,9 +1387,9 @@ def setup_calendar_routes() -> APIRouter:
         if tz_hint:
             set_user_tz_name(tz_hint)
 
-        url, model, headers = resolve_endpoint("utility")
+        url, model, headers = resolve_endpoint("utility", owner=owner or None)
         if not url:
-            url, model, headers = resolve_endpoint("default")
+            url, model, headers = resolve_endpoint("default", owner=owner or None)
         if not url or not model:
             return {"ok": False, "error": "No LLM endpoint configured"}
 

routes/chat_helpers.py

@@ -75,7 +75,7 @@ def _enforce_chat_privileges(request, sess) -> None:
     allowlist, or HTTPException(429) if the user has hit their daily message
     cap. No-op for unauthenticated callers or when auth_manager is absent
     (single-user mode). Admins receive ADMIN_PRIVILEGES from get_privileges,
-    which means empty allowed_models / zero cap → no-op for them.
+    which means unrestricted allowed_models / zero cap -> no-op for them.
     """
     try:
         user = get_current_user(request)
@@ -88,8 +88,18 @@ def _enforce_chat_privileges(request, sess) -> None:
         return
 
     privs = auth_manager.get_privileges(user) or {}
-    allowed = privs.get("allowed_models") or []
-    if allowed and sess.model and sess.model not in allowed:
+
+    # Explicit "block everything" sentinel takes precedence over the
+    # allowlist — it's the only way to distinguish "user clicked [None]"
+    # (block all) from "user clicked [All]" (no restriction), since both
+    # otherwise produce an empty `allowed_models` list.
+    if privs.get("block_all_models"):
+        raise HTTPException(403, f"Your account is not allowed to use model '{sess.model}'.")
+
+    allowed_raw = privs.get("allowed_models")
+    allowed = allowed_raw if isinstance(allowed_raw, list) else []
+    restricted = bool(privs.get("allowed_models_restricted")) or bool(allowed)
+    if restricted and sess.model and sess.model not in allowed:
         raise HTTPException(403, f"Your account is not allowed to use model '{sess.model}'.")
 
     cap = int(privs.get("max_messages_per_day") or 0)
@@ -194,14 +204,26 @@ def try_fallback_endpoint(sess, session_id: str) -> dict | None:
     Returns {"model": ..., "endpoint_url": ..., "endpoint_name": ...} or None.
     """
     import requests as _req
-    from src.endpoint_resolver import build_chat_url, build_headers, build_models_url, normalize_base
+    from src.endpoint_resolver import (
+        build_chat_url,
+        build_headers,
+        build_models_url,
+        normalize_base,
+        resolve_endpoint_runtime,
+    )
+    from src.chatgpt_subscription import is_chatgpt_subscription_base
 
     current_url = sess.endpoint_url or ""
+    owner = getattr(sess, "owner", None)
     db = SessionLocal()
     try:
-        endpoints = db.query(ModelEndpoint).filter(
+        q = db.query(ModelEndpoint).filter(
             ModelEndpoint.is_enabled == True
-        ).all()
+        )
+        if owner:
+            from src.auth_helpers import owner_filter
+            q = owner_filter(q, ModelEndpoint, owner)
+        endpoints = q.all()
     finally:
         db.close()
 
@@ -210,26 +232,33 @@ def try_fallback_endpoint(sess, session_id: str) -> dict | None:
         # Skip current endpoint
         if current_url and base in current_url:
             continue
-        # Quick ping
-        ping_url = build_models_url(base)
-        headers = build_headers(ep.api_key, base)
         try:
-            r = _req.get(ping_url, headers=headers, timeout=5)
-            r.raise_for_status()
-            data = r.json()
-            models = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
-            if not models:
-                models = [
-                    m.get("name") or m.get("model")
-                    for m in (data.get("models") or [])
-                    if m.get("name") or m.get("model")
-                ]
+            base, api_key = resolve_endpoint_runtime(ep, owner=owner)
+        except Exception:
+            continue
+        ping_url = build_models_url(base)
+        headers = build_headers(api_key, base)
+        try:
+            if ping_url:
+                r = _req.get(ping_url, headers=headers, timeout=5)
+                r.raise_for_status()
+                data = r.json()
+                models = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
+                if not models:
+                    models = [
+                        m.get("name") or m.get("model")
+                        for m in (data.get("models") or [])
+                        if m.get("name") or m.get("model")
+                    ]
+            else:
+                models = json.loads(ep.cached_models or "[]")
             if not models:
                 continue
             # Found a working endpoint — update session
             new_model = models[0]
             chat_url = build_chat_url(base)
-            new_headers = build_headers(ep.api_key, base)
+            new_headers = build_headers(api_key, base)
+            persisted_headers = {} if is_chatgpt_subscription_base(base) else new_headers
 
             sess.model = new_model
             sess.endpoint_url = chat_url
@@ -241,7 +270,7 @@ def try_fallback_endpoint(sess, session_id: str) -> dict | None:
                 _db.query(DBSession).filter(DBSession.id == session_id).update({
                     "model": new_model,
                     "endpoint_url": chat_url,
-                    "headers": json.dumps(new_headers),
+                    "headers": persisted_headers,
                 })
                 _db.commit()
             finally:
@@ -275,11 +304,16 @@ def extract_preset(chat_handler, preset_id) -> PresetInfo:
 async def preprocess(
     chat_handler, message, att_ids, sess,
     auto_opened_docs: Optional[list] = None,
+    allow_tool_preprocessing: bool = True,
 ) -> PreprocessedMessage:
     """Run chat_handler.preprocess_message and wrap the result."""
     enhanced, user_content, text_ctx, yt_transcripts, att_meta = (
         await chat_handler.preprocess_message(
-            message, att_ids, sess, auto_opened_docs=auto_opened_docs
+            message,
+            att_ids,
+            sess,
+            auto_opened_docs=auto_opened_docs,
+            allow_tool_preprocessing=allow_tool_preprocessing,
         )
     )
     return PreprocessedMessage(
@@ -329,16 +363,26 @@ def _session_url_matches_endpoint(session_url: str, endpoint_base: str) -> bool:
         return False
 
 
+def _has_auth_keys(headers) -> bool:
+    """True if a headers dict carries an Authorization/x-api-key entry."""
+    return isinstance(headers, dict) and any(
+        k.lower() in ('authorization', 'x-api-key') for k in headers
+    )
+
+
 def resolve_session_auth(sess, session_id: str, owner: Optional[str] = None):
     """Ensure session has auth headers — resolve from endpoint DB if missing."""
-    has_auth = sess.headers and isinstance(sess.headers, dict) and any(
-        k.lower() in ('authorization', 'x-api-key') for k in sess.headers
-    )
-    if has_auth:
+    try:
+        from src.chatgpt_subscription import is_chatgpt_subscription_base
+        is_chatgpt_subscription = is_chatgpt_subscription_base(getattr(sess, "endpoint_url", "") or "")
+    except Exception:
+        is_chatgpt_subscription = False
+    has_auth = _has_auth_keys(sess.headers)
+    if has_auth and not is_chatgpt_subscription:
         return
 
     try:
-        from src.endpoint_resolver import build_headers, normalize_base
+        from src.endpoint_resolver import build_headers, resolve_endpoint_runtime
         db = SessionLocal()
         try:
             target_url = getattr(sess, "endpoint_url", "") or ""
@@ -354,10 +398,30 @@ def resolve_session_auth(sess, session_id: str, owner: Optional[str] = None):
             for ep in q.all():
                 if not _session_url_matches_endpoint(target_url, ep.base_url or ""):
                     continue
-                if not ep.api_key:
+                try:
+                    base, api_key = resolve_endpoint_runtime(ep, owner=owner)
+                except Exception as e:
+                    logger.warning("Failed to resolve provider auth for session %s: %s", session_id, e)
+                    return
+                if not api_key:
+                    # No usable key (e.g. ChatGPT Subscription needs re-auth).
+                    return
+                sess.headers = build_headers(api_key, base)
+                if is_chatgpt_subscription:
+                    # The bearer is short-lived and re-resolved per request, so it
+                    # stays request-local and is never written to the plaintext
+                    # sessions.headers column. Proactively strip any bearer an
+                    # older code path may have persisted so it does not linger.
+                    stale_q = db.query(DBSession).filter(DBSession.id == session_id)
+                    if owner:
+                        stale_q = stale_q.filter(DBSession.owner == owner)
+                    stored = stale_q.first()
+                    if stored is not None and _has_auth_keys(stored.headers):
+                        stale_q.update({"headers": {}})
+                        db.commit()
+                        logger.info(f"Cleared persisted ChatGPT Subscription bearer from session {session_id}")
+                    logger.debug(f"Resolved request-local ChatGPT Subscription auth for session {session_id}")
                     return
-                base = normalize_base(ep.base_url or "")
-                sess.headers = build_headers(ep.api_key, base)
                 update_q = db.query(DBSession).filter(DBSession.id == session_id)
                 if owner:
                     update_q = update_q.filter(DBSession.owner == owner)
@@ -401,7 +465,12 @@ def _normalize_model_id_from_cache(sess) -> Optional[str]:
 
     db = SessionLocal()
     try:
-        endpoints = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True).all()
+        q = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True)
+        owner = getattr(sess, "owner", None)
+        if owner:
+            from src.auth_helpers import owner_filter
+            q = owner_filter(q, ModelEndpoint, owner)
+        endpoints = q.all()
         for ep in endpoints:
             try:
                 if normalize_base(getattr(ep, "base_url", "") or "") != session_base:
@@ -448,6 +517,7 @@ async def build_chat_context(
     webhook_manager=None,
     use_enhanced_message: bool = False,
     agent_mode: bool = False,
+    allow_tool_preprocessing: bool = True,
 ) -> ChatContext:
     """Build the full context (preface + messages) for an LLM call.
 
@@ -465,6 +535,7 @@ async def build_chat_context(
     preprocessed = await preprocess(
         chat_handler, message, att_ids or [], sess,
         auto_opened_docs=auto_opened_docs,
+        allow_tool_preprocessing=allow_tool_preprocessing,
     )
 
     # Add user message to history
@@ -483,6 +554,9 @@ async def build_chat_context(
     # Skills injection respects its own enable toggle (mirrors memory_enabled).
     # When off, the "Available skills" index is not added to the prompt.
     skills_enabled = not incognito and uprefs.get("skills_enabled", True)
+    if not allow_tool_preprocessing:
+        mem_enabled = False
+        skills_enabled = False
     logger.debug(
         "Memory enabled=%s for user=%s (incognito=%s, no_memory=%s, pref=%s)",
         mem_enabled, user, incognito, no_memory, uprefs.get("memory_enabled", "NOT_SET"),
@@ -490,11 +564,11 @@ async def build_chat_context(
 
     # Use RAG?
     use_rag_val = (str(use_rag).lower() != "false") if use_rag is not None else True
-    if incognito:
+    if incognito or not allow_tool_preprocessing:
         use_rag_val = False
 
     # If pre-fetched search context was provided (compare mode), skip live web search
-    skip_web = bool(search_context)
+    skip_web = bool(search_context) or not allow_tool_preprocessing
 
     # Build context preface
     # The stream path uses enhanced_message (with CoT/preprocessing applied),
@@ -521,7 +595,7 @@ async def build_chat_context(
     used_memories = getattr(chat_processor, '_last_used_memories', [])
 
     # Inject pre-fetched search context (compare mode)
-    if search_context:
+    if search_context and allow_tool_preprocessing:
         preface.append(untrusted_context_message("prefetched search context", search_context))
 
     # YouTube transcripts
@@ -530,16 +604,40 @@ async def build_chat_context(
 
     # Normalize model ID. Prefer cached endpoint models so group chat does not
     # re-hit slow local /models endpoints on every participant turn.
-    norm = _normalize_model_id_from_cache(sess) or normalize_model_id(sess.endpoint_url, sess.model)
+    norm = _normalize_model_id_from_cache(sess) or normalize_model_id(
+        sess.endpoint_url,
+        sess.model,
+        owner=getattr(sess, "owner", None),
+    )
     if norm:
         sess.model = norm
 
     # Build messages
     messages = preface + sess.get_context_messages()
 
+    # Current date/time — injected as a standalone *user*-role context message
+    # placed immediately before the latest user turn, NOT folded into the
+    # system prompt. Its text changes every minute, and local OpenAI-compatible
+    # backends (llama.cpp / LM Studio) key their KV-cache prefix off the
+    # system message byte-for-byte; mixing ever-changing timestamp text into
+    # it would invalidate the cached prefix on every request (issue #2927).
+    # Placing it at the tail also keeps it out of the stable
+    # preface+history prefix, so that prefix stays byte-identical turn over
+    # turn (modulo the genuinely new history entries) and the cache survives.
+    if not agent_mode:
+        try:
+            from src.user_time import current_datetime_context_message
+            _dt_msg = current_datetime_context_message()
+            if messages and messages[-1].get("role") == "user":
+                messages.insert(len(messages) - 1, _dt_msg)
+            else:
+                messages.append(_dt_msg)
+        except Exception:
+            logger.debug("Failed to add current date/time context", exc_info=True)
+
     # Auto-compact
     messages, context_length, was_compacted = await maybe_compact(
-        sess, sess.endpoint_url, sess.model, messages, sess.headers,
+        sess, sess.endpoint_url, sess.model, messages, sess.headers, owner=user,
     )
     messages = trim_for_context(messages, context_length)
 
@@ -772,7 +870,19 @@ def save_assistant_response(
 ):
     """Add assistant response to session history. In incognito mode, keeps in-memory context but skips DB persistence."""
     md = dict(last_metrics) if last_metrics else {}
-    md["model"] = sess.model
+    def _model_value(value) -> str:
+        if value is None:
+            return ""
+        if not isinstance(value, str):
+            value = str(value)
+        return value.strip()
+
+    requested_model = _model_value(md.get("requested_model") or md.get("selected_model") or getattr(sess, "model", ""))
+    actual_model = _model_value(md.get("model") or md.get("actual_model") or requested_model)
+    if requested_model:
+        md["requested_model"] = requested_model
+    if actual_model:
+        md["model"] = actual_model
     if character_name:
         md["character_name"] = character_name
     if web_sources:
@@ -821,6 +931,54 @@ def save_assistant_response(
     return None
 
 
+def _is_session_stream_active(session_id: str) -> bool:
+    """Best-effort check for "is a chat completion currently streaming for
+    this session?" — used to keep background extraction from overlapping a
+    main completion and competing for the local backend's processing slots
+    (issue #2927). Lazily imports the route module's live registry to avoid
+    a circular import (chat_routes imports this module at load time)."""
+    try:
+        from routes import chat_routes as _cr
+        return session_id in getattr(_cr, "_active_streams", {})
+    except Exception:
+        return False
+
+
+async def _run_extraction_jobs_sequentially(session_id: str, jobs: list, max_wait_s: float = 120.0):
+    """Run queued background-extraction coroutines one at a time, only once
+    no chat completion is actively streaming for this session.
+
+    As diagnosed in issue #2927, firing memory/skill extraction concurrently
+    with the main chat completion (or with each other) makes them compete for
+    the local backend's limited processing slots, evicting the main
+    conversation's cached KV-cache checkpoint and forcing a full prompt
+    re-evaluation on the next turn. Waiting for the stream to go idle and then
+    running the jobs strictly in sequence keeps at most one "side" request in
+    flight against the backend at any time, and never alongside the user's
+    own conversation.
+    """
+    # Wait for the triggering turn's own stream to finish winding down (it
+    # almost always already has by the time this task gets scheduled — this
+    # is a small safety margin, not the primary mechanism).
+    waited = 0.0
+    poll = 0.25
+    while _is_session_stream_active(session_id) and waited < max_wait_s:
+        await asyncio.sleep(poll)
+        waited += poll
+
+    for name, job in jobs:
+        # Re-check before each job: a fast follow-up message from the user
+        # may have started a new stream for this session while we waited.
+        waited = 0.0
+        while _is_session_stream_active(session_id) and waited < max_wait_s:
+            await asyncio.sleep(poll)
+            waited += poll
+        try:
+            await job
+        except Exception:
+            logger.warning("[bg-extract] %s extraction job failed for session %s", name, session_id, exc_info=True)
+
+
 def run_post_response_tasks(
     sess,
     session_manager,
@@ -841,21 +999,37 @@ def run_post_response_tasks(
     skills_manager=None,
     owner: str = None,
     extract_skills: bool = True,
+    allow_background_extraction: bool = True,
 ):
-    """Fire background tasks after a completed response: memory extraction, webhooks, auto-name, skill extraction."""
+    """Fire background tasks after a completed response: memory extraction, webhooks, auto-name, skill extraction.
+
+    Memory/skill extraction are queued to run *sequentially*, after the main
+    completion stream for this session has fully wound down — never
+    concurrently with it or with each other. As diagnosed in issue #2927,
+    firing these "side" LLM calls in parallel with the main chat completion
+    makes them compete for the local backend's limited processing slots
+    (llama.cpp defaults to 4), evicting the main conversation's cached
+    checkpoint and forcing a full prompt re-evaluation on the next turn. By
+    the time this function runs the main response is already saved, but the
+    extraction calls themselves are still async — queuing them through
+    ``_queue_background_extraction`` keeps them from overlapping the *next*
+    turn's request too.
+    """
+    _extraction_jobs: list = []
+
     # Memory extraction — only every 4th message pair to avoid excess LLM calls
     _msg_count = len(sess.history) if hasattr(sess, 'history') else 0
     _should_extract = (_msg_count >= 4) and (_msg_count % 4 == 0)
-    if not incognito and not compare_mode and _should_extract and uprefs.get("auto_memory", True):
+    if allow_background_extraction and not incognito and not compare_mode and _should_extract and uprefs.get("auto_memory", True):
         from services.memory.memory_extractor import extract_and_store
         from src.task_endpoint import resolve_task_endpoint
         t_url, t_model, t_headers = resolve_task_endpoint(
             sess.endpoint_url, sess.model, sess.headers, owner=owner,
         )
-        asyncio.create_task(extract_and_store(
+        _extraction_jobs.append(("memory", extract_and_store(
             sess, memory_manager, memory_vector,
             t_url, t_model, t_headers,
-        ))
+        )))
 
     # Skill extraction from complex agent runs. Only when the user actually
     # chose agent mode — not a chat we auto-escalated for a notes/calendar
@@ -873,6 +1047,7 @@ def run_post_response_tasks(
     )
     if (
         extract_skills
+        and allow_background_extraction
         and auto_skills_enabled
         and not incognito
         and not compare_mode
@@ -890,12 +1065,15 @@ def run_post_response_tasks(
                 sess.endpoint_url, sess.model, sess.headers, owner=owner,
             )
             logger.debug("[skill-extract] dispatching extractor (model=%s)", s_model)
-            asyncio.create_task(maybe_extract_skill(
+            _extraction_jobs.append(("skill", maybe_extract_skill(
                 sess, skills_manager,
                 s_url, s_model, s_headers,
                 agent_rounds, agent_tool_calls,
                 owner=owner,
-            ))
+            )))
+
+    if _extraction_jobs:
+        asyncio.create_task(_run_extraction_jobs_sequentially(session_id, _extraction_jobs))
 
     # Token accumulation
     if last_metrics:

routes/chat_routes.py

@@ -20,6 +20,7 @@ from src import agent_runs
 from src.model_context import estimate_tokens
 from src.chat_helpers import coerce_message_and_session
 from src.endpoint_resolver import normalize_base as _normalize_base, build_chat_url
+from src.session_search import search_session_messages
 from src.prompt_security import untrusted_context_message
 from core.exceptions import SessionNotFoundError
 from src.auth_helpers import get_current_user
@@ -39,6 +40,7 @@ from routes.chat_helpers import (
     _enforce_chat_privileges,
 )
 from src.action_intents import classify_tool_intent as _classify_tool_intent
+from src.tool_policy import build_effective_tool_policy
 
 logger = logging.getLogger(__name__)
 
@@ -60,6 +62,33 @@ def _stream_set(session_id: str, **fields) -> None:
     rec.update(fields)
 
 
+def _resolve_request_workspace(request, raw_value) -> tuple:
+    """Resolve the posted workspace for this request: (workspace, rejected).
+
+    Privilege is checked BEFORE the path ever touches the filesystem. Only
+    admin/single-user callers can use the workspace-backed file/shell tools,
+    so only they get vet_workspace() and the workspace_rejected signal. For
+    any other caller the submitted value is dropped uniformly, with no vetting
+    and no event: otherwise the presence/absence of workspace_rejected would
+    let a non-admin chat caller probe which host paths exist.
+
+    vet_workspace rejects non-directories, sensitive roots (.ssh, .gnupg,
+    ...), and filesystem roots; on rejection there is no confinement and the
+    default tool-path allowlist applies. The rejected value is surfaced so the
+    stream can tell an admin client (which believes a workspace is active)
+    that it was dropped.
+    """
+    requested = (raw_value or "").strip()
+    if not requested:
+        return "", ""
+    from src.tool_security import owner_is_admin_or_single_user
+    if not owner_is_admin_or_single_user(get_current_user(request)):
+        return "", ""
+    from src.tool_execution import vet_workspace
+    workspace = vet_workspace(requested) or ""
+    return workspace, (requested if not workspace else "")
+
+
 def _session_url_matches_endpoint(session_url: str, endpoint_base: str) -> bool:
     if not session_url or not endpoint_base:
         return False
@@ -167,13 +196,20 @@ def _recover_empty_session_model(sess, session_id: str, owner: str | None = None
     Covers the window between endpoint setup and the first chat send: the
     picker showed a model in the dropdown but the session record never got
     written (Issue #587 — UI uses the cached endpoint list, not s.model).
-    Without this, we'd POST the upstream with model="" and get a generic
-    401/503 instead of using the model the user already picked.
-
-    Returns True iff sess.model was repaired.
+    For ChatGPT Subscription, also repairs stale OpenAI API model names such as
+    ``gpt-5`` that are not accepted by the Codex-backed ChatGPT account route.
     """
-    if getattr(sess, "model", None):
-        return False
+    current_model = (getattr(sess, "model", "") or "").strip()
+    endpoint_url = (getattr(sess, "endpoint_url", "") or "").strip()
+    is_chatgpt_subscription = False
+    if current_model:
+        try:
+            from src.chatgpt_subscription import is_chatgpt_subscription_base
+            is_chatgpt_subscription = is_chatgpt_subscription_base(endpoint_url)
+            if not is_chatgpt_subscription:
+                return False
+        except Exception:
+            return False
     db = SessionLocal()
     try:
         # Prefer the endpoint whose base URL matches the session — we know the
@@ -192,16 +228,51 @@ def _recover_empty_session_model(sess, session_id: str, owner: str | None = None
                     break
         if not ep:
             return False
+        if not is_chatgpt_subscription:
+            try:
+                from src.chatgpt_subscription import is_chatgpt_subscription_base
+                is_chatgpt_subscription = is_chatgpt_subscription_base(getattr(ep, "base_url", "") or endpoint_url)
+            except Exception:
+                is_chatgpt_subscription = False
         try:
             cached = json.loads(ep.cached_models) if isinstance(ep.cached_models, str) else (ep.cached_models or [])
         except Exception:
             cached = []
         if not cached:
+            visible = []
+        else:
+            try:
+                visible = _visible_models(cached, getattr(ep, "hidden_models", None))
+            except Exception:
+                visible = cached
+        if current_model and current_model in {str(item).strip() for item in visible}:
             return False
-        try:
-            visible = _visible_models(cached, getattr(ep, "hidden_models", None))
-        except Exception:
-            visible = cached
+        if is_chatgpt_subscription:
+            live_models = []
+            if getattr(ep, "provider_auth_id", None):
+                try:
+                    from src.chatgpt_subscription import fetch_available_models
+                    from src.endpoint_resolver import resolve_endpoint_runtime
+                    _base, api_key = resolve_endpoint_runtime(ep, owner=owner)
+                    if api_key:
+                        live_models = fetch_available_models(api_key)
+                        if live_models:
+                            ep.cached_models = json.dumps(live_models)
+                            db.commit()
+                except Exception:
+                    live_models = []
+            # ChatGPT Subscription recovery must use the live Codex catalog.
+            # Cached rows are only trusted above to avoid revalidating a model
+            # that is already present in the visible picker list.
+            cached = live_models
+            if not cached:
+                return False
+            try:
+                visible = _visible_models(cached, getattr(ep, "hidden_models", None))
+            except Exception:
+                visible = cached
+            if current_model and current_model in {str(item).strip() for item in visible}:
+                return False
         if not visible:
             return False
         model = visible[0]
@@ -211,14 +282,17 @@ def _recover_empty_session_model(sess, session_id: str, owner: str | None = None
         # Persist so the next request, websocket reconnect, or page reload
         # picks up the same model (we'd otherwise re-pick on every send
         # and silently switch on the user if the cached order shifts).
-        db_session = db.query(DBSession).filter(DBSession.id == session_id).first()
+        db_session_q = db.query(DBSession).filter(DBSession.id == session_id)
+        if owner:
+            db_session_q = db_session_q.filter(DBSession.owner == owner)
+        db_session = db_session_q.first()
         if db_session:
             db_session.model = model
             db_session.updated_at = datetime.utcnow()
             db.commit()
         sess.model = model
         logger.info(
-            "Recovered empty session model for %s — picked %r from endpoint %s",
+            "Recovered session model for %s — picked %r from endpoint %s",
             session_id, model, ep.id,
         )
         return True
@@ -304,8 +378,13 @@ def setup_chat_routes(
         # non-streaming path can't be used to bypass).
         _enforce_chat_privileges(request, sess)
 
+        tool_policy = build_effective_tool_policy(last_user_message=message)
+        allow_tool_preprocessing = not tool_policy.block_all_tool_calls
+
         # Inline memory command
-        memory_response = await chat_handler.handle_memory_command(sess, message)
+        memory_response = None
+        if not tool_policy.blocks("manage_memory"):
+            memory_response = await chat_handler.handle_memory_command(sess, message)
         if memory_response:
             return {"response": memory_response}
 
@@ -319,10 +398,15 @@ def setup_chat_routes(
             use_web=use_web,
             time_filter=time_filter,
             webhook_manager=webhook_manager,
+            allow_tool_preprocessing=allow_tool_preprocessing,
         )
 
         # Research injection
-        if use_research:
+        research_blocked_by_policy = (
+            tool_policy.blocks("trigger_research")
+            or tool_policy.blocks("manage_research")
+        )
+        if use_research and not research_blocked_by_policy:
             try:
                 _r_ep, _r_model, _r_headers = _resolve_research_endpoint(sess)
                 research_ctx = await research_handler.call_research_service(
@@ -343,6 +427,7 @@ def setup_chat_routes(
             temperature=ctx.preset.temperature,
             max_tokens=ctx.preset.max_tokens,
             prompt_type=preset_id,
+            session_id=session,
         )
         _clean_reply, _clean_md = clean_thinking_for_save(reply, {"model": sess.model})
         sess.add_message(ChatMessage("assistant", _clean_reply, metadata=_clean_md))
@@ -357,6 +442,7 @@ def setup_chat_routes(
             ctx.uprefs, memory_manager, memory_vector, webhook_manager,
             character_name=ctx.preset.character_name,
             owner=ctx.user,
+            allow_background_extraction=not tool_policy.block_all_tool_calls,
         )
 
         return {"response": reply}
@@ -388,19 +474,34 @@ def setup_chat_routes(
         use_research = form_data.get("use_research")
         time_filter = form_data.get("time_filter")
         preset_id = form_data.get("preset_id")
-        allow_bash = form_data.get("allow_bash")
-        allow_web_search = form_data.get("allow_web_search")
+        # Issue #3229: API callers send JSON, not FormData.  Read from the
+        # JSON body as fallback so callers who send {"allow_bash": true}
+        # actually get bash enabled.
+        allow_bash = form_data.get("allow_bash") or (body or {}).get("allow_bash")
+        allow_web_search = form_data.get("allow_web_search") or (body or {}).get("allow_web_search")
         use_rag = form_data.get("use_rag")
         search_context = form_data.get("search_context")  # pre-fetched web search results (compare mode)
         compare_mode = str(form_data.get("compare_mode", "")).lower() == "true"
         incognito = str(form_data.get("incognito", "")).lower() == "true"
+        # Plan mode is not part of the merge-ready UI. Ignore stale clients or
+        # manual form posts that still send plan_mode=true.
+        plan_mode = False
         chat_mode = str(form_data.get("mode", "")).lower()  # 'chat' or 'agent'
-        # Workspace: confine the agent's file/shell tools to this folder. Validate
-        # it's a real directory; ignore (no confinement) otherwise.
-        workspace = (form_data.get("workspace") or "").strip()
-        if workspace:
-            _ws_real = os.path.realpath(os.path.expanduser(workspace))
-            workspace = _ws_real if os.path.isdir(_ws_real) else ""
+        # Workspace: confine the agent's file/shell tools to this folder.
+        workspace, workspace_rejected = _resolve_request_workspace(
+            request, form_data.get("workspace")
+        )
+        # Plan mode is a modifier on agent mode — it only makes sense with tools.
+        if plan_mode:
+            chat_mode = "agent"
+        # An approved plan being EXECUTED: the frontend sends the checklist back
+        # on each turn so we can pin it in context. This way a long plan on a
+        # weak model survives history truncation — the agent can always re-read
+        # the plan. Ignored while still proposing (plan_mode on). Capped so a
+        # huge plan can't blow the prompt.
+        approved_plan = ""
+        if not plan_mode:
+            approved_plan = (form_data.get("approved_plan") or "").strip()[:8192]
         # Did the USER explicitly pick agent mode? (vs. us auto-escalating
         # below). Skill extraction should only learn from real agent sessions,
         # not chats we quietly promoted for a notes/calendar intent.
@@ -479,11 +580,6 @@ def setup_chat_routes(
                 do_research = True
                 logger.info(f"Session {session} in research_pending — auto-triggering research")
 
-        # Persist session mode (research > agent > chat)
-        _effective_mode = 'research' if do_research else (chat_mode or 'chat')
-        if _effective_mode in ('agent', 'research', 'chat'):
-            set_session_mode(session, _effective_mode)
-
         att_ids = []
         if body and isinstance(body.get("attachments"), list):
             att_ids = [str(x) for x in body["attachments"]]
@@ -494,6 +590,10 @@ def setup_chat_routes(
                 pass
 
         no_memory = str(form_data.get("no_memory", "")).lower() == "true"
+        pre_context_tool_policy = build_effective_tool_policy(
+            last_user_message=message,
+        )
+        allow_tool_preprocessing = not pre_context_tool_policy.block_all_tool_calls
 
         # Build shared context (stream path uses enhanced_message for context preface)
         ctx = await build_chat_context(
@@ -515,6 +615,7 @@ def setup_chat_routes(
             # manage_skills (agent mode). In plain chat or incognito the
             # index would be useless / unwanted noise.
             agent_mode=(chat_mode == "agent"),
+            allow_tool_preprocessing=allow_tool_preprocessing,
         )
 
         _research_flags = {"do": do_research}  # Mutable container for generator scope
@@ -568,7 +669,7 @@ def setup_chat_routes(
             # leak a doc that belongs to a DIFFERENT session.
             if not active_doc:
                 try:
-                    from src.tool_implementations import get_active_document
+                    from src.agent_tools.document_tools import get_active_document
                     _mem_id = get_active_document()
                     if _mem_id:
                         _mem_q = _doc_db.query(DBDocument).filter(DBDocument.id == _mem_id)
@@ -589,9 +690,13 @@ def setup_chat_routes(
 
         # Build disabled-tools set from frontend toggles + user privileges
         disabled_tools = set()
-        if str(allow_bash).lower() != "true":
+        # Only disable bash/web_search when the caller *explicitly* set them
+        # to a falsy value.  When unset (None), defer to per-user privilege
+        # checks below — this lets admins with can_use_bash=True use bash
+        # by default without having to send allow_bash in every request.
+        if allow_bash is not None and str(allow_bash).lower() != "true":
             disabled_tools.add("bash")
-        if str(allow_web_search).lower() != "true":
+        if allow_web_search is not None and str(allow_web_search).lower() != "true":
             disabled_tools.add("web_search")
             disabled_tools.add("web_fetch")
 
@@ -659,6 +764,32 @@ def setup_chat_routes(
             if chat_mode == 'chat':
                 disabled_tools.update({"bash", "python", "read_file", "write_file", "web_search", "web_fetch", "search_chats", "manage_tasks"})
 
+        # Plan mode: investigate read-only, propose a plan, don't mutate. Block
+        # every tool not on the read-only allowlist. (stream_agent_loop enforces
+        # this again + drops MCP, so this is belt-and-suspenders.)
+        if plan_mode:
+            from src.tool_security import plan_mode_disabled_tools
+            disabled_tools.update(plan_mode_disabled_tools())
+
+        tool_policy = build_effective_tool_policy(
+            disabled_tools=disabled_tools,
+            last_user_message=message,
+        )
+        disabled_tools = tool_policy.all_disabled_names()
+        research_blocked_by_policy = bool(
+            tool_policy.blocks("trigger_research")
+            or tool_policy.blocks("manage_research")
+        )
+        effective_do_research = bool(
+            do_research and _research_flags["do"] and not research_blocked_by_policy
+        )
+
+        # Persist session mode after policy/privilege gates so blocked research
+        # turns remain ordinary chat/agent streams and saved messages.
+        _effective_mode = 'research' if effective_do_research else (chat_mode or 'chat')
+        if _effective_mode in ('agent', 'research', 'chat'):
+            set_session_mode(session, _effective_mode)
+
         async def stream_with_save() -> AsyncGenerator[str, None]:
             # _effective_mode is read-only here; closure captures it from
             # the outer scope. (Was `nonlocal` but never reassigned.)
@@ -666,7 +797,14 @@ def setup_chat_routes(
             web_sources = ctx.web_sources
 
             # Register active stream for partial-save safety net
-            _active_streams[session] = {"status": "streaming", "partial": "", "query": message, "is_research": do_research, "mode": _effective_mode}
+            _active_streams[session] = {"status": "streaming", "partial": "", "query": message, "is_research": effective_do_research, "mode": _effective_mode}
+
+            # The client sent a workspace the server refused to bind (deleted
+            # folder, file path, sensitive dir, filesystem root). Tell it up
+            # front so the UI can clear the pill instead of displaying a
+            # confinement that is not actually in effect.
+            if workspace_rejected:
+                yield f"data: {json.dumps({'type': 'workspace_rejected', 'data': {'path': workspace_rejected}})}\n\n"
 
             if ctx.preprocessed.attachment_meta:
                 yield f"data: {json.dumps({'type': 'attachments', 'data': ctx.preprocessed.attachment_meta})}\n\n"
@@ -690,7 +828,7 @@ def setup_chat_routes(
                 yield f"data: {json.dumps({'type': 'memories_used', 'data': ctx.used_memories})}\n\n"
 
             # Run research as a background task (survives page refresh)
-            if do_research and _research_flags["do"]:
+            if effective_do_research:
                 _r_ep, _r_model, _r_headers = _resolve_research_endpoint(sess)
                 _auth_keys = list(_r_headers.keys()) if _r_headers else []
                 logger.info(f"Research endpoint resolved: model={_r_model}, endpoint={_r_ep}, auth_keys={_auth_keys}, sess_headers_keys={list(sess.headers.keys()) if isinstance(sess.headers, dict) else type(sess.headers)}")
@@ -829,7 +967,7 @@ def setup_chat_routes(
                 _fallback_candidates = []
 
             # Send model name early so the frontend can show it during streaming
-            _model_suffix = "Research" if do_research else None
+            _model_suffix = "Research" if effective_do_research else None
             _model_info = {"type": "model_info", "model": sess.model}
             if _model_suffix:
                 _model_info["suffix"] = _model_suffix
@@ -839,6 +977,12 @@ def setup_chat_routes(
 
             if _is_image_generation_session(sess, owner=_user):
                 from src.settings import get_setting
+                if tool_policy.blocks("generate_image"):
+                    _blocked_msg = tool_policy.reason_for("generate_image")
+                    yield f'data: {json.dumps({"delta": _blocked_msg})}\n\n'
+                    yield "data: [DONE]\n\n"
+                    _active_streams.pop(session, None)
+                    return
                 if not get_setting("image_gen_enabled", True):
                     yield f'data: {json.dumps({"delta": "Image generation is disabled by the administrator."})}\n\n'
                     yield "data: [DONE]\n\n"
@@ -873,6 +1017,8 @@ def setup_chat_routes(
             elif chat_mode == "chat":
                 _chat_start = time.time()
                 _answered_by = None  # set if the selected model failed and a fallback answered
+                _requested_model = sess.model
+                _actual_model = None
                 # ── Chat mode: call stream_llm directly, NO tools, NO document access ──
                 try:
                     _chat_candidates = [(sess.endpoint_url, sess.model, sess.headers)] + _fallback_candidates
@@ -888,6 +1034,7 @@ def setup_chat_routes(
                         max_tokens=ctx.preset.max_tokens,
                         prompt_type=preset_id,
                         tools=None,
+                        session_id=session,
                     ):
                         if chunk.startswith("data: ") and not chunk.startswith("data: [DONE]"):
                             try:
@@ -905,10 +1052,18 @@ def setup_chat_routes(
                                     # Selected model failed; a fallback answered.
                                     # Forward the notice and remember the real model.
                                     _answered_by = data.get("answered_by") or _answered_by
+                                    _actual_model = _actual_model or _answered_by
+                                    data["selected_model"] = data.get("selected_model") or _requested_model
                                     yield chunk
+                                elif data.get("type") == "model_actual":
+                                    _actual_model = data.get("model") or _actual_model
+                                    data["requested_model"] = _requested_model
+                                    yield f'data: {json.dumps(data)}\n\n'
                                 elif data.get("type") == "usage":
                                     last_metrics = data.get("data", {})
-                                    last_metrics["model"] = _answered_by or sess.model
+                                    _reported_model = last_metrics.get("model")
+                                    last_metrics["requested_model"] = _requested_model
+                                    last_metrics["model"] = _reported_model or _actual_model or _answered_by or _requested_model
                                     if ctx.context_length and last_metrics.get("input_tokens"):
                                         pct = min(round((last_metrics["input_tokens"] / ctx.context_length) * 100, 1), 100.0)
                                         last_metrics["context_percent"] = pct
@@ -945,7 +1100,8 @@ def setup_chat_routes(
                                     "tokens_per_second": _tps,
                                     "context_percent": _ctx_pct,
                                     "context_length": ctx.context_length,
-                                    "model": sess.model,
+                                    "model": _actual_model or _answered_by or _requested_model,
+                                    "requested_model": _requested_model,
                                     "usage_source": "estimated",
                                 }
                                 yield f'data: {json.dumps({"type": "metrics", "data": last_metrics})}\n\n'
@@ -957,7 +1113,7 @@ def setup_chat_routes(
                                     rag_sources=ctx.rag_sources,
                                     research_sources=research_sources,
                                     used_memories=ctx.used_memories,
-                                    do_research=do_research,
+                                    do_research=effective_do_research,
                                     incognito=incognito,
                                 )
                                 if _saved_id:
@@ -967,14 +1123,22 @@ def setup_chat_routes(
                                     last_metrics, ctx.uprefs, memory_manager, memory_vector, webhook_manager,
                                     incognito=incognito, compare_mode=compare_mode,
                                     character_name=ctx.preset.character_name,
-                                                            owner=_user,
+                                    owner=_user,
+                                    allow_background_extraction=not tool_policy.block_all_tool_calls,
                                 )
                             _stream_set(session, status="done")
                             yield chunk
                 except (asyncio.CancelledError, GeneratorExit):
                     if full_response:
                         logger.info("Client disconnected mid-stream (chat mode) for session %s, saving partial (%d chars)", session, len(full_response))
-                        _stopped_content, _stopped_md = clean_thinking_for_save(full_response, {"stopped": True, "model": sess.model})
+                        _stopped_content, _stopped_md = clean_thinking_for_save(
+                            full_response,
+                            {
+                                "stopped": True,
+                                "model": _actual_model or _answered_by or _requested_model,
+                                "requested_model": _requested_model,
+                            },
+                        )
                         sess.add_message(ChatMessage("assistant", _stopped_content, metadata=_stopped_md))
                         if not incognito:
                             session_manager.save_sessions()
@@ -986,6 +1150,8 @@ def setup_chat_routes(
                 _agent_rounds = 0
                 _agent_tool_calls = 0
                 _answered_by = None  # set if the selected model failed and a fallback answered
+                _requested_model = sess.model
+                _actual_model = None
                 try:
                     from src.settings import get_setting
                     from src.agent_tools import MAX_AGENT_ROUNDS as _DEFAULT_ROUNDS
@@ -1012,8 +1178,11 @@ def setup_chat_routes(
                         active_document=active_doc,
                         session_id=session,
                         disabled_tools=disabled_tools if disabled_tools else None,
+                        tool_policy=tool_policy,
                         owner=_user,
                         fallbacks=_fallback_candidates,
+                        plan_mode=plan_mode,
+                        approved_plan=approved_plan or None,
                         workspace=workspace or None,
                     ):
                         if chunk.startswith("data: ") and not chunk.startswith("data: [DONE]"):
@@ -1035,6 +1204,8 @@ def setup_chat_routes(
                                     "doc_stream_open", "doc_stream_delta",
                                     "doc_update", "doc_suggestions", "ui_control",
                                     "rounds_exhausted",
+                                    "ask_user",
+                                    "plan_update",
                                 ):
                                     if data.get("type") == "agent_step":
                                         _agent_rounds = max(_agent_rounds, data.get("round", 1))
@@ -1047,10 +1218,18 @@ def setup_chat_routes(
                                     # model so metrics reflect it, not the masked
                                     # selected model.
                                     _answered_by = data.get("answered_by") or _answered_by
+                                    _actual_model = _actual_model or _answered_by
+                                    data["selected_model"] = data.get("selected_model") or _requested_model
                                     yield chunk
+                                elif data.get("type") == "model_actual":
+                                    _actual_model = data.get("model") or _actual_model
+                                    data["requested_model"] = _requested_model
+                                    yield f'data: {json.dumps(data)}\n\n'
                                 elif data.get("type") == "metrics":
                                     last_metrics = data.get("data", {})
-                                    last_metrics["model"] = _answered_by or sess.model
+                                    _reported_model = last_metrics.get("model")
+                                    last_metrics["requested_model"] = last_metrics.get("requested_model") or _requested_model
+                                    last_metrics["model"] = _reported_model or _actual_model or _answered_by or _requested_model
                                     yield f'data: {json.dumps({"type": "metrics", "data": last_metrics})}\n\n'
                             except json.JSONDecodeError:
                                 yield chunk
@@ -1078,6 +1257,7 @@ def setup_chat_routes(
                                     skills_manager=skills_manager,
                                     owner=_user,
                                     extract_skills=user_requested_agent,
+                                    allow_background_extraction=not tool_policy.block_all_tool_calls,
                                 )
                             _stream_set(session, status="done")
                             yield chunk
@@ -1091,7 +1271,14 @@ def setup_chat_routes(
                     try:
                         if full_response:
                             logger.info("Client disconnected mid-stream for session %s, saving partial response (%d chars)", session, len(full_response))
-                            _stopped_content2, _stopped_md2 = clean_thinking_for_save(full_response, {"stopped": True, "model": sess.model})
+                            _stopped_content2, _stopped_md2 = clean_thinking_for_save(
+                                full_response,
+                                {
+                                    "stopped": True,
+                                    "model": _actual_model or _answered_by or _requested_model,
+                                    "requested_model": _requested_model,
+                                },
+                            )
                             sess.add_message(ChatMessage("assistant", _stopped_content2, metadata=_stopped_md2))
                             if not incognito:
                                 session_manager.save_sessions()
@@ -1110,11 +1297,29 @@ def setup_chat_routes(
             finally:
                 _active_streams.pop(session, None)
 
-        # Run the stream as a DETACHED background task so it survives the client
-        # closing the tab / navigating away (true terminal-agent behavior). The
-        # SSE response just subscribes (replay buffered output + live); dropping
-        # the SSE only removes a subscriber — the run keeps going and saves the
-        # assistant message on completion regardless. Reconnect via /api/chat/resume.
+        # Compare panes are short-lived, single-shot generations whose sessions
+        # exist only to drive that one pane — there's nothing to "resume" and
+        # the user expects the pane's Stop button (which aborts the fetch,
+        # closing this SSE) to promptly cancel the upstream LLM call. Detaching
+        # them would keep burning upstream tokens/compute after the pane is
+        # stopped or the comparison is abandoned, and would surface a stale
+        # "still streaming" /resume target for a session nobody will revisit.
+        #
+        # So: stream them directly (no agent_runs wrapping). Starlette cancels
+        # the underlying async generator (raising CancelledError/GeneratorExit
+        # inside it) as soon as it notices the client disconnected — which the
+        # mode-specific except blocks above already handle by saving the
+        # partial response exactly once. This stops the upstream call promptly
+        # without waiting on the next streamed chunk.
+        #
+        # Normal chat/agent streams keep the DETACHED behavior below: they
+        # survive the client closing the tab / navigating away. The SSE response just subscribes (replay
+        # buffered output + live); dropping the SSE only removes a subscriber —
+        # the run keeps going and saves the assistant message on completion
+        # regardless. Reconnect via /api/chat/resume.
+        if compare_mode:
+            return StreamingResponse(_safe_stream(), media_type="text/event-stream")
+
         agent_runs.start(session, _safe_stream())
         return StreamingResponse(agent_runs.subscribe(session), media_type="text/event-stream")
 
@@ -1185,45 +1390,16 @@ def setup_chat_routes(
             return []
 
         _user = get_current_user(request)
-        query_term = q.strip()
-        db = SessionLocal()
-        try:
-            base_q = (
-                db.query(DBChatMessage, DBSession.name)
-                .join(DBSession, DBChatMessage.session_id == DBSession.id)
-                .filter(
-                    DBSession.archived == False,
-                    DBChatMessage.content.ilike(f"%{query_term}%"),
-                    DBChatMessage.role.in_(["user", "assistant"]),
-                )
+        return [
+            result.to_dict()
+            for result in search_session_messages(
+                q,
+                limit=limit,
+                owner=_user,
+                restrict_owner=_user is not None,
+                include_legacy_owner=False,
             )
-            if _user:
-                base_q = base_q.filter(DBSession.owner == _user)
-            rows = base_q.order_by(DBChatMessage.timestamp.desc()).limit(limit).all()
-
-            results = []
-            for msg, session_name in rows:
-                content = msg.content or ""
-                lower_content = content.lower()
-                idx = lower_content.find(query_term.lower())
-                if idx == -1:
-                    snippet = content[:120]
-                else:
-                    start = max(0, idx - 50)
-                    end = min(len(content), idx + len(query_term) + 50)
-                    snippet = ("..." if start > 0 else "") + content[start:end] + ("..." if end < len(content) else "")
-
-                results.append({
-                    "session_id": msg.session_id,
-                    "session_name": session_name or "Untitled",
-                    "role": msg.role,
-                    "content_snippet": snippet,
-                    "timestamp": msg.timestamp.isoformat() if msg.timestamp else None,
-                })
-
-            return results
-        finally:
-            db.close()
+        ]
 
     # ------------------------------------------------------------------ #
     # POST /api/rewrite — lightweight rewrite of last AI message (no tools)

routes/chatgpt_subscription_routes.py

@@ -0,0 +1,170 @@
+"""ChatGPT Subscription device-flow setup routes."""
+
+import json
+import logging
+import uuid
+from typing import Dict, Optional
+
+from fastapi import HTTPException, Request
+
+from core.database import ModelEndpoint, ProviderAuthSession, SessionLocal, utcnow_naive
+from routes.device_flow import (
+    DeviceFlowPoll,
+    DeviceFlowStart,
+    PendingDeviceFlowStore,
+    create_device_flow_router,
+)
+from src.auth_helpers import get_current_user
+from src import chatgpt_subscription
+
+logger = logging.getLogger(__name__)
+
+_DEVICE_FLOW_STORE = PendingDeviceFlowStore()
+
+
+def _provision_endpoint(tokens: Dict, owner: Optional[str]) -> Dict:
+    access_token = tokens.get("access_token")
+    refresh_token = tokens.get("refresh_token")
+    if not access_token or not refresh_token:
+        raise ValueError("ChatGPT token response was missing access_token or refresh_token")
+
+    base = chatgpt_subscription.DEFAULT_CHATGPT_SUBSCRIPTION_BASE_URL
+    models = chatgpt_subscription.fetch_available_models(access_token)
+    if not models:
+        raise ValueError("ChatGPT Subscription connected, but no usable Codex models were discovered for this account.")
+    db = SessionLocal()
+    try:
+        auth = (
+            db.query(ProviderAuthSession)
+            .filter(
+                ProviderAuthSession.provider == chatgpt_subscription.CHATGPT_SUBSCRIPTION_PROVIDER,
+                ProviderAuthSession.owner == owner,
+            )
+            .first()
+        )
+        if auth is None:
+            auth = ProviderAuthSession(
+                id=str(uuid.uuid4())[:8],
+                provider=chatgpt_subscription.CHATGPT_SUBSCRIPTION_PROVIDER,
+                owner=owner,
+                label="ChatGPT Subscription",
+                base_url=base,
+                auth_mode="chatgpt",
+            )
+            db.add(auth)
+        auth.base_url = base
+        auth.access_token = access_token
+        auth.refresh_token = refresh_token
+        auth.last_refresh = utcnow_naive()
+        auth.auth_mode = "chatgpt"
+
+        ep = (
+            db.query(ModelEndpoint)
+            .filter(
+                ModelEndpoint.base_url == base,
+                ModelEndpoint.provider_auth_id == auth.id,
+                ModelEndpoint.owner == owner,
+            )
+            .first()
+        )
+        if ep is None:
+            ep = ModelEndpoint(
+                id=str(uuid.uuid4())[:8],
+                name="ChatGPT Subscription",
+                base_url=base,
+                model_type="llm",
+                endpoint_kind="api",
+                owner=owner,
+            )
+            db.add(ep)
+        ep.name = "ChatGPT Subscription"
+        ep.base_url = base
+        ep.api_key = None
+        ep.provider_auth_id = auth.id
+        ep.is_enabled = True
+        ep.supports_tools = False
+        ep.model_type = "llm"
+        ep.endpoint_kind = "api"
+        ep.model_refresh_mode = "manual"
+        ep.cached_models = json.dumps(models)
+        db.commit()
+        result = {
+            "id": ep.id,
+            "name": ep.name,
+            "base_url": ep.base_url,
+            "models": models,
+        }
+    finally:
+        db.close()
+
+    try:
+        from routes.model_routes import _invalidate_models_cache
+
+        _invalidate_models_cache()
+    except Exception:
+        pass
+    return result
+
+
+def _start_device_flow(request: Request, _form) -> DeviceFlowStart:
+    try:
+        data = chatgpt_subscription.request_device_code()
+    except Exception as exc:
+        raise chatgpt_subscription.to_http_exception(exc)
+
+    device_auth_id = data.get("device_auth_id")
+    user_code = data.get("user_code")
+    if not device_auth_id or not user_code:
+        raise HTTPException(502, "ChatGPT did not return a complete device code")
+    verification_uri = data.get("verification_uri") or f"{chatgpt_subscription.CHATGPT_OAUTH_ISSUER}/codex/device"
+    return DeviceFlowStart(
+        pending={
+            "device_auth_id": device_auth_id,
+            "user_code": user_code,
+            "owner": get_current_user(request) or None,
+        },
+        response={
+            "user_code": user_code,
+            "verification_uri": verification_uri,
+        },
+        interval=int(data.get("interval") or 5),
+        expires_in=int(data.get("expires_in") or 900),
+    )
+
+
+def _poll_device_flow(_request: Request, pending: Dict) -> DeviceFlowPoll:
+    try:
+        data = chatgpt_subscription.poll_device_auth(pending["device_auth_id"], pending["user_code"])
+    except Exception as exc:
+        logger.debug("ChatGPT device poll failed: %s", exc)
+        return DeviceFlowPoll.pending(str(exc))
+
+    authorization_code = data.get("authorization_code")
+    code_verifier = data.get("code_verifier")
+    if authorization_code and code_verifier:
+        try:
+            tokens = chatgpt_subscription.exchange_authorization_code(authorization_code, code_verifier)
+            result = _provision_endpoint(tokens, pending["owner"])
+        except Exception as exc:
+            logger.exception("ChatGPT Subscription endpoint provisioning failed")
+            raise chatgpt_subscription.to_http_exception(exc)
+        return DeviceFlowPoll.authorized(result)
+
+    err = data.get("error") or data.get("status")
+    if err in ("authorization_pending", "pending", None):
+        return DeviceFlowPoll.pending()
+    if err == "slow_down":
+        return DeviceFlowPoll.slow_down(int(data.get("interval") or 0) or None)
+    if err in ("expired_token", "access_denied", "denied"):
+        return DeviceFlowPoll.failed(err)
+    return DeviceFlowPoll.pending(err or "unknown")
+
+
+def setup_chatgpt_subscription_routes():
+    return create_device_flow_router(
+        prefix="/api/chatgpt-subscription",
+        tags=["chatgpt-subscription"],
+        store=_DEVICE_FLOW_STORE,
+        start_flow=_start_device_flow,
+        poll_flow=_poll_device_flow,
+    )

routes/codex_routes.py

@@ -15,10 +15,13 @@ from typing import Any
 from fastapi import APIRouter, BackgroundTasks, Body, HTTPException, Request
 from fastapi.responses import StreamingResponse
 
-from src.auth_helpers import require_user
+from src.auth_helpers import require_authenticated_request, require_user
 from src.tool_implementations import do_manage_notes
+from src.constants import COOKBOOK_STATE_FILE
 
 
+COOKBOOK_READ_SCOPES = {"cookbook:read", "cookbook:launch"}
+COOKBOOK_LAUNCH_SCOPES = {"cookbook:launch"}
 TODO_READ_SCOPES = {"todos:read", "todos:write"}
 TODO_WRITE_SCOPES = {"todos:write"}
 EMAIL_READ_SCOPES = {"email:read", "email:draft", "email:send"}
@@ -39,7 +42,9 @@ async def _as_owner(request: Request, owner: str, fn, *args, **kwargs):
     the scope-gated owner (not the "api" pseudo-user the bearer middleware sets).
     Restores the original value when done. Works for sync and async handlers."""
     orig = getattr(request.state, "current_user", None)
+    orig_api_token = getattr(request.state, "api_token", None)
     request.state.current_user = owner
+    request.state.api_token = False
     try:
         result = fn(*args, **kwargs)
         if asyncio.iscoroutine(result):
@@ -47,6 +52,13 @@ async def _as_owner(request: Request, owner: str, fn, *args, **kwargs):
         return result
     finally:
         request.state.current_user = orig
+        if orig_api_token is None:
+            try:
+                delattr(request.state, "api_token")
+            except AttributeError:
+                pass
+        else:
+            request.state.api_token = orig_api_token
 
 
 def _scope_owner(request: Request, allowed: set[str]) -> str:
@@ -130,6 +142,11 @@ def setup_codex_routes(
                     "actions": ["library", "read", "create", "delete"],
                     "available": documents_library_endpoint is not None,
                 },
+                "cookbook": {
+                    "read": scoped(COOKBOOK_READ_SCOPES),
+                    "launch": scoped(COOKBOOK_LAUNCH_SCOPES),
+                    "actions": ["tasks", "servers", "output", "serve", "stop"],
+                },
             },
             "safety": {
                 "email_send_requires_confirmation": True,
@@ -139,7 +156,7 @@ def setup_codex_routes(
 
     @router.get("/plugin.zip")
     def plugin_zip(request: Request):
-        require_user(request)
+        require_authenticated_request(request)
         root = Path(__file__).resolve().parent.parent / "integrations" / "codex"
         if not root.exists():
             raise HTTPException(404, "Codex plugin bundle not found")
@@ -373,6 +390,374 @@ def setup_codex_routes(
             raise HTTPException(400, f"Invalid document payload: {exc}")
         return await _as_owner(request, owner, documents_create_endpoint, request, req)
 
+    # ── Cookbook surface ──
+    # Lets the agent run the same launch / monitor / kill loop the user
+    # would do by hand in the Cookbook UI: read the current task list +
+    # tmux output, launch a serve task, stop one.  Two scopes:
+    #   cookbook:read   — list tasks + tail output + list servers
+    #   cookbook:launch — also start/stop serves (host shell exec)
+    # `cookbook:launch` is genuinely powerful: /api/model/serve runs SSH'd
+    # commands on the user's hosts. The existing _validate_serve_cmd
+    # allowlist (vllm/python3/sglang/llama-server/etc., no shell metachars)
+    # keeps the agent inside the same sandbox the UI uses.
+
+    async def _run_shell(cmd: str, timeout: float = 15.0) -> dict:
+        """Run a shell command, return {exit_code, stdout, stderr}."""
+        import asyncio as _asyncio
+        try:
+            proc = await _asyncio.create_subprocess_shell(
+                cmd,
+                stdout=_asyncio.subprocess.PIPE,
+                stderr=_asyncio.subprocess.PIPE,
+            )
+            try:
+                stdout_b, stderr_b = await _asyncio.wait_for(proc.communicate(), timeout=timeout)
+            except _asyncio.TimeoutError:
+                proc.kill()
+                return {"exit_code": -1, "stdout": "", "stderr": "timed out"}
+            return {
+                "exit_code": proc.returncode,
+                "stdout": stdout_b.decode(errors="replace"),
+                "stderr": stderr_b.decode(errors="replace"),
+            }
+        except Exception as exc:
+            return {"exit_code": -1, "stdout": "", "stderr": str(exc)}
+
+    def _read_cookbook_state() -> dict:
+        from pathlib import Path as _Path
+        import json as _json
+        p = _Path(COOKBOOK_STATE_FILE)
+        if not p.exists():
+            return {}
+        try:
+            return _json.loads(p.read_text(encoding="utf-8"))
+        except Exception:
+            return {}
+
+    def _redact_task(t: dict) -> dict:
+        """Strip secrets before returning to the agent."""
+        clean = {k: v for k, v in t.items() if k not in ("hf_token", "_secrets")}
+        if isinstance(clean.get("payload"), dict):
+            pl = clean["payload"]
+            clean["payload"] = {k: v for k, v in pl.items()
+                                if k not in ("hf_token", "_secrets")}
+        return clean
+
+    @router.get("/cookbook/tasks")
+    async def codex_cookbook_tasks(request: Request):
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
+        state = _read_cookbook_state()
+        tasks = state.get("tasks") or []
+        return {"tasks": [_redact_task(t) for t in tasks]}
+
+    @router.get("/cookbook/servers")
+    async def codex_cookbook_servers(request: Request):
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
+        state = _read_cookbook_state()
+        servers = state.get("env", {}).get("servers") or []
+        # Strip ssh creds / passwords; keep only what's needed to pick a host.
+        cleaned = []
+        for s in servers:
+            cleaned.append({
+                "name": s.get("name"),
+                "host": s.get("host"),
+                "port": s.get("port"),
+                "env": s.get("env"),
+                "envPath": s.get("envPath"),
+                "platform": s.get("platform"),
+                "modelDirs": s.get("modelDirs"),
+            })
+        return {"servers": cleaned}
+
+    @router.get("/cookbook/output/{session_id}")
+    async def codex_cookbook_output(request: Request, session_id: str, tail: int = 400):
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
+        # Defensive: session_id must be the tmux-style id we issue
+        # (`serve-XXXX` / `cookbook-XXXX` / `queue-XXXX`); anything else
+        # would let the agent run arbitrary `tmux capture-pane` targets.
+        import re as _re
+        if not _re.fullmatch(r"[a-zA-Z0-9_-]+", session_id):
+            raise HTTPException(400, "Invalid session id")
+        tail = max(20, min(int(tail or 400), 4000))
+        # Resolve the task's host (if any) from cookbook state so we can
+        # ssh to the right box, exactly as the UI does in _reconnectTask.
+        state = _read_cookbook_state()
+        tasks = state.get("tasks") or []
+        task = next((t for t in tasks if t.get("sessionId") == session_id), None)
+        if task is None:
+            raise HTTPException(404, "task not found")
+        host = (task.get("remoteHost") or "").strip()
+        ssh_port = (task.get("sshPort") or "").strip()
+        # Prefer the persisted log file over the tmux pane. The pane gets
+        # overwritten by the post-crash neofetch banner + bash prompt the
+        # moment vllm exits; the log file is the raw stdout/stderr and
+        # survives unchanged. Falls back to pane for older tasks predating
+        # the tee-to-log runner change.
+        log_path = f"/tmp/odysseus-tmux/{session_id}.log"
+        inner = (
+            f"if [ -s {log_path} ]; then tail -n {tail} {log_path}; "
+            f"else tmux capture-pane -t {session_id} -p -S -{tail}; fi"
+        )
+        if host:
+            port_flag = f"-p {ssh_port} " if ssh_port and ssh_port != "22" else ""
+            import shlex
+            cmd = f"ssh {port_flag}{host} {shlex.quote(inner)}"
+        else:
+            cmd = inner
+        result = await _run_shell(cmd, timeout=15)
+        return {
+            "session_id": session_id,
+            "host": host or "local",
+            "exit_code": result.get("exit_code"),
+            "output": result.get("stdout", ""),
+            "task": _redact_task(task),
+        }
+
+    @router.post("/cookbook/serve")
+    async def codex_cookbook_serve(request: Request, body: dict[str, Any] = Body(default_factory=dict)):
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
+        # Wraps /api/model/serve with the SAME validation the UI uses.
+        # _validate_serve_cmd (called inside model_serve) rejects shell
+        # metachars and requires the leading binary to be in the
+        # cookbook allowlist (vllm / python3 / sglang / llama-server / ...).
+        from routes.cookbook_helpers import ServeRequest
+        # Accept friendly aliases agents naturally reach for. Without these,
+        # passing `host` silently maps to nothing and the serve runs LOCAL
+        # instead of on the intended remote — exactly the bug an agent
+        # would never debug on its own.
+        norm = dict(body or {})
+        if "host" in norm and "remote_host" not in norm:
+            norm["remote_host"] = norm.pop("host")
+        if "model" in norm and "repo_id" not in norm:
+            norm["repo_id"] = norm.pop("model")
+        if "ssh_port" not in norm and "port" in norm and (str(norm.get("port") or "").isdigit() and int(norm["port"]) >= 1000):
+            # Heuristic: if `port` looks like an SSH port (≥1000) and there's
+            # no explicit ssh_port, treat it as such. UI ports (8000, 8001,
+            # 30000) belong inside the cmd string, not here.
+            pass  # leave as-is — user's `port` here is ambiguous; skip remap.
+        try:
+            req = ServeRequest(**norm)
+        except Exception as exc:
+            raise HTTPException(400, f"Invalid serve payload: {exc}")
+        serve_endpoint = _find_endpoint(None, "POST", "/api/model/serve")
+        # Fall back to importing from the cookbook router registered on app.
+        if serve_endpoint is None:
+            from fastapi import FastAPI
+            app: FastAPI = request.app
+            for route in app.routes:
+                if getattr(route, "path", None) == "/api/model/serve" and "POST" in getattr(route, "methods", set()):
+                    serve_endpoint = route.endpoint
+                    break
+        if serve_endpoint is None:
+            raise HTTPException(503, "model serve endpoint unavailable")
+        return await serve_endpoint(request, req)
+
+    @router.post("/cookbook/stop/{session_id}")
+    async def codex_cookbook_stop(request: Request, session_id: str):
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
+        import re as _re
+        if not _re.fullmatch(r"[a-zA-Z0-9_-]+", session_id):
+            raise HTTPException(400, "Invalid session id")
+        state = _read_cookbook_state()
+        tasks = state.get("tasks") or []
+        task = next((t for t in tasks if t.get("sessionId") == session_id), None)
+        host = ((task or {}).get("remoteHost") or "").strip()
+        ssh_port = ((task or {}).get("sshPort") or "").strip()
+        if host:
+            port_flag = f"-p {ssh_port} " if ssh_port and ssh_port != "22" else ""
+            cmd = f"ssh {port_flag}{host} \"tmux kill-session -t {session_id}\""
+        else:
+            cmd = f"tmux kill-session -t {session_id}"
+        result = await _run_shell(cmd, timeout=10)
+        return {"session_id": session_id, "exit_code": result.get("exit_code"), "host": host or "local"}
+
+    @router.get("/cookbook/cached")
+    async def codex_cookbook_cached(request: Request, host: str | None = None):
+        """List cached models on a configured server (or local if host is omitted).
+        Mirrors `list_cached_models` from the chat agent so external agents have
+        the same inventory view before deciding what to serve/download."""
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
+        # Hit /api/model/cached internally, with the same modelDirs the chat
+        # agent's list_cached_models would resolve from cookbook state.
+        state = _read_cookbook_state()
+        env = state.get("env") if isinstance(state, dict) else {}
+        servers = (env.get("servers") if isinstance(env, dict) else None) or []
+        HF_DEFAULTS = {"~/.cache/huggingface/hub", "~/.cache/huggingface"}
+        def _dirs_for(srv: dict) -> str:
+            mds = srv.get("modelDirs") if isinstance(srv, dict) else None
+            if isinstance(mds, list):
+                extras = [d for d in mds if isinstance(d, str) and d.strip() and d.strip() not in HF_DEFAULTS]
+                return ",".join(extras)
+            if isinstance(mds, str) and mds.strip() not in HF_DEFAULTS:
+                return mds
+            return ""
+        # Resolve friendly host name → real host (matches list_cached_models flow).
+        resolved_host = host or ""
+        srv: dict[str, Any] = {}
+        if host:
+            srv = next(
+                (s for s in servers if isinstance(s, dict)
+                 and (s.get("name") == host or s.get("host") == host)),
+                {},
+            )
+            if srv and srv.get("host"):
+                resolved_host = srv["host"]
+        else:
+            srv = next((s for s in servers if isinstance(s, dict) and not (s.get("host") or "").strip()), {})
+        params: dict[str, str] = {}
+        if resolved_host:
+            params["host"] = resolved_host
+        md = _dirs_for(srv)
+        if md:
+            params["model_dir"] = md
+        if srv.get("port"):
+            params["ssh_port"] = str(srv["port"])
+        if srv.get("platform"):
+            params["platform"] = srv["platform"]
+        cached_endpoint = _find_endpoint(None, "GET", "/api/model/cached")
+        if cached_endpoint is None:
+            from fastapi import FastAPI
+            app: FastAPI = request.app
+            for route in app.routes:
+                if getattr(route, "path", None) == "/api/model/cached" and "GET" in getattr(route, "methods", set()):
+                    cached_endpoint = route.endpoint
+                    break
+        if cached_endpoint is None:
+            raise HTTPException(503, "model cached endpoint unavailable")
+        # The endpoint reads host/model_dir/ssh_port/platform as kwargs.
+        return await cached_endpoint(
+            request,
+            host=params.get("host") or None,
+            model_dir=params.get("model_dir") or None,
+            ssh_port=params.get("ssh_port") or None,
+            platform=params.get("platform") or None,
+        )
+
+    @router.get("/cookbook/presets")
+    async def codex_cookbook_presets(request: Request):
+        """List saved serve presets (model + host + port + launch cmd).
+        Counterpart to `list_serve_presets`. Use BEFORE composing a `serve`
+        body — the user's saved preset usually has the working cmd already."""
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
+        state = _read_cookbook_state()
+        presets = state.get("presets") or []
+        out = []
+        for p in presets:
+            if not isinstance(p, dict):
+                continue
+            out.append({
+                "name": p.get("name"),
+                "model": p.get("model") or p.get("modelId"),
+                "host": p.get("host") or p.get("remoteHost"),
+                "port": p.get("port"),
+                "cmd": p.get("cmd"),
+            })
+        return {"presets": out, "default_host": (state.get("env") or {}).get("defaultServer", "")}
+
+    @router.post("/cookbook/preset/{name}")
+    async def codex_cookbook_serve_preset(request: Request, name: str):
+        """Launch a saved preset by name. Reuses the working cmd + host the
+        user already saved, avoiding the cmd-allowlist trial-and-error loop."""
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
+        import re as _re
+        if not _re.fullmatch(r"[A-Za-z0-9 _.:@\-]+", name):
+            raise HTTPException(400, "Invalid preset name")
+        state = _read_cookbook_state()
+        presets = state.get("presets") or []
+        lname = name.lower().strip()
+        chosen = next(
+            (p for p in presets if isinstance(p, dict) and (p.get("name") or "").lower() == lname),
+            None,
+        )
+        if chosen is None:
+            chosen = next(
+                (p for p in presets if isinstance(p, dict) and lname in (p.get("name") or "").lower()),
+                None,
+            )
+        if chosen is None:
+            raise HTTPException(404, f"No preset matching {name!r}")
+        repo_id = chosen.get("model") or chosen.get("modelId") or ""
+        cmd = (chosen.get("cmd") or "").strip()
+        host = chosen.get("host") or chosen.get("remoteHost") or ""
+        if not repo_id or not cmd or cmd.startswith("(adopted"):
+            raise HTTPException(400, f"Preset {chosen.get('name')!r} has no launchable cmd "
+                                     "(adopted from external launch). Use POST /cookbook/serve "
+                                     "with the actual cmd instead.")
+        # Reuse the serve handler we already validated.
+        from routes.cookbook_helpers import ServeRequest
+        body = {"repo_id": repo_id, "cmd": cmd}
+        if host:
+            body["remote_host"] = host
+        try:
+            req = ServeRequest(**body)
+        except Exception as exc:
+            raise HTTPException(400, f"Preset payload invalid: {exc}")
+        serve_endpoint = _find_endpoint(None, "POST", "/api/model/serve")
+        if serve_endpoint is None:
+            from fastapi import FastAPI
+            app: FastAPI = request.app
+            for route in app.routes:
+                if getattr(route, "path", None) == "/api/model/serve" and "POST" in getattr(route, "methods", set()):
+                    serve_endpoint = route.endpoint
+                    break
+        if serve_endpoint is None:
+            raise HTTPException(503, "model serve endpoint unavailable")
+        return await serve_endpoint(request, req)
+
+    @router.post("/cookbook/adopt")
+    async def codex_cookbook_adopt(request: Request, body: dict[str, Any] = Body(default_factory=dict)):
+        """Adopt an existing tmux session (one started via raw ssh+tmux) into
+        cookbook tracking. Needed when serve_model rejects a cmd and the
+        agent falls back to direct ssh — without adoption the session is
+        invisible to the UI. Body: {tmux_session, model, host?, port?}."""
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
+        norm = dict(body or {})
+        sess = (norm.get("tmux_session") or norm.get("session_id") or "").strip()
+        model = (norm.get("model") or norm.get("repo_id") or "").strip()
+        host = (norm.get("host") or norm.get("remote_host") or "").strip()
+        port = norm.get("port") or 8000
+        import re as _re
+        if not sess or not _re.fullmatch(r"[a-zA-Z0-9_-]+", sess):
+            raise HTTPException(400, "tmux_session required, [a-zA-Z0-9_-]+ only")
+        if not model:
+            raise HTTPException(400, "model required")
+        # Verify the tmux session exists on the target host before adopting.
+        import shlex
+        if host:
+            check = f"ssh {shlex.quote(host)} 'tmux has-session -t {shlex.quote(sess)}'"
+        else:
+            check = f"tmux has-session -t {shlex.quote(sess)}"
+        chk = await _run_shell(check, timeout=8)
+        if chk.get("exit_code") not in (0, None):
+            raise HTTPException(404, f"tmux session {sess!r} not found on {host or 'local'}")
+        # Write into cookbook_state.json.
+        import time as _t, json as _json
+        from core.atomic_io import atomic_write_json
+        from pathlib import Path as _Path
+        cookbook_state_path = _Path(COOKBOOK_STATE_FILE)
+        try:
+            state = _json.loads(cookbook_state_path.read_text(encoding="utf-8"))
+        except Exception:
+            state = {}
+        tasks = state.setdefault("tasks", [])
+        if any(isinstance(t, dict) and t.get("sessionId") == sess for t in tasks):
+            return {"ok": True, "already_tracked": True, "session_id": sess}
+        tasks.append({
+            "id": sess, "sessionId": sess,
+            "name": model.split("/")[-1] if "/" in model else model,
+            "type": "serve", "status": "running",
+            "output": f"Adopted externally-launched session {sess!r} on {host or 'local'}.",
+            "ts": int(_t.time() * 1000),
+            "payload": {"repo_id": model, "remote_host": host, "_cmd": "(adopted — launched outside cookbook)", "port": int(port)},
+            "remoteHost": host, "sshPort": "", "platform": "linux",
+            "_serveReady": False, "_endpointAdded": False, "_adoptedExternally": True,
+        })
+        try:
+            atomic_write_json(cookbook_state_path, state)
+        except Exception as exc:
+            raise HTTPException(500, f"state write failed: {exc}")
+        return {"ok": True, "session_id": sess, "host": host or "local"}
+
     return router
 
 
@@ -387,7 +772,7 @@ def setup_claude_routes() -> APIRouter:
 
     @router.get("/plugin.zip")
     def plugin_zip(request: Request):
-        require_user(request)
+        require_authenticated_request(request)
         # Only ship the skills/ subtree so extracting at ~/.claude/ doesn't dump
         # README.md or other bundle metadata into the user's claude config dir.
         skills_root = Path(__file__).resolve().parent.parent / "integrations" / "claude" / "skills"

routes/compare_routes.py

@@ -12,6 +12,7 @@ import logging
 from core.database import Comparison, SessionLocal
 from core.session_manager import SessionManager
 from src.auth_helpers import get_current_user
+from routes.session_routes import _reject_raw_endpoint_url_for_non_admin
 
 logger = logging.getLogger(__name__)
 
@@ -38,6 +39,24 @@ def _owned_endpoint_by_url(db, base_url, owner):
     return owner_filter(q, ModelEndpoint, owner).first()
 
 
+def _owned_endpoint_by_id(db, endpoint_id, owner):
+    """ModelEndpoint whose id == `endpoint_id` and is VISIBLE to `owner` (their
+    own rows + legacy null-owner "shared" rows); None otherwise.
+
+    Preferred over _owned_endpoint_by_url for credential resolution: two visible
+    endpoints can share the same base_url but hold DIFFERENT api_keys (e.g. two
+    accounts on the same provider). A base_url-only match returns whichever row
+    sorts first, so it can copy the WRONG owner-scoped key into the [CMP] session.
+    An id pins the exact registered endpoint, so /api/compare/start prefers it and
+    only falls back to URL matching for legacy / admin raw-URL callers. Owner
+    scoping is identical to _owned_endpoint_by_url (a null/empty owner is a no-op).
+    """
+    from core.database import ModelEndpoint
+    from src.auth_helpers import owner_filter
+    q = db.query(ModelEndpoint).filter(ModelEndpoint.id == endpoint_id)
+    return owner_filter(q, ModelEndpoint, owner).first()
+
+
 class RecordVoteRequest(BaseModel):
     prompt: str
     models: List[str]
@@ -54,8 +73,10 @@ def setup_compare_routes(session_manager: SessionManager):
         prompt: str = Form(...),
         model_a: str = Form(...),
         model_b: str = Form(...),
-        endpoint_a: str = Form(...),
-        endpoint_b: str = Form(...),
+        endpoint_a: str = Form(""),
+        endpoint_b: str = Form(""),
+        endpoint_a_id: str = Form(""),
+        endpoint_b_id: str = Form(""),
         is_blind: str = Form("true"),
     ):
         """Create two ephemeral sessions and a comparison record.
@@ -63,10 +84,10 @@ def setup_compare_routes(session_manager: SessionManager):
         Returns the comparison ID and the two session IDs so the client
         can fire two independent SSE streams to /api/chat_stream.
         """
+        user = getattr(request.state, 'current_user', None)
         comp_id = str(uuid.uuid4())
         sid_a = str(uuid.uuid4())
         sid_b = str(uuid.uuid4())
-        user = getattr(request.state, 'current_user', None)
 
         # Blind mapping: randomly assign left/right
         blind = str(is_blind).lower() == "true"
@@ -87,31 +108,94 @@ def setup_compare_routes(session_manager: SessionManager):
         # de-anonymizing the comparison before the user votes (issue #1285).
         slot_name = {session_left: "Model A", session_right: "Model B"}
 
-        # Create ephemeral sessions (prefixed [CMP])
-        for sid, model, endpoint in [(sid_a, model_a, endpoint_a), (sid_b, model_b, endpoint_b)]:
+        # SECURITY: resolve and validate BOTH endpoints before creating any
+        # session. Compare copies a registered endpoint's Authorization header
+        # into the [CMP] session, so validating one endpoint while creating its
+        # session, then rejecting the other, would leave a partial compare
+        # session behind with that header attached. Doing all the owner-scope
+        # resolution + raw-URL rejection up front means a 403 on either endpoint
+        # aborts the whole request with nothing created and no header copied.
+        from src.endpoint_resolver import build_chat_url, build_headers, normalize_base
+        resolved = []
+        db = SessionLocal()
+        try:
+            for sid, model, endpoint, endpoint_id in [
+                (sid_a, model_a, endpoint_a, endpoint_a_id),
+                (sid_b, model_b, endpoint_b, endpoint_b_id),
+            ]:
+                # Prefer an explicit endpoint id: it pins the EXACT registered
+                # endpoint (and its api_key), even when two endpoints visible to
+                # the caller share a base_url with different keys — a URL-only
+                # match would copy whichever row sorts first, i.e. possibly the
+                # wrong key. Fall back to URL resolution only for legacy / admin
+                # raw-URL callers that don't send an id.
+                eid = endpoint_id.strip() if isinstance(endpoint_id, str) else ""
+                if eid:
+                    ep = _owned_endpoint_by_id(db, eid, user)
+                    if ep is None:
+                        # An id the caller can't see (wrong owner / deleted) must
+                        # NOT silently fall back to a same-URL row with a different
+                        # key — that's exactly the mix-up ids exist to prevent.
+                        raise HTTPException(404, "Model endpoint not found")
+                    # The id already resolved the endpoint; ignore any raw URL the
+                    # caller also sent and dial the stored config instead.
+                    endpoint = ep.base_url
+                elif not endpoint:
+                    raise HTTPException(
+                        422, "endpoint_a/endpoint_b or endpoint_a_id/endpoint_b_id is required"
+                    )
+                else:
+                    # Resolve the supplied URL to a ModelEndpoint the caller owns
+                    # (their own rows + legacy null-owner shared rows), scoped so a
+                    # comparison can't borrow another user's private endpoint key.
+                    base = normalize_base(endpoint)
+                    ep = _owned_endpoint_by_url(db, base, user)
+                # Reject *unregistered* raw URLs for signed-in non-admins; a
+                # matched registered endpoint supplies an id so the caller can
+                # still compare endpoints they own. Blanket-rejecting here (the
+                # earlier `endpoint_id=None` call) locked non-admins out of
+                # compare entirely, since compare resolves endpoints by URL with
+                # no endpoint_id. Mirrors the gallery inpaint/harmonize checks.
+                # Raised here (phase 1), before any session exists.
+                _reject_raw_endpoint_url_for_non_admin(
+                    request, user, str(ep.id) if ep is not None else None, endpoint
+                )
+                # Bind the [CMP] session to the RESOLVED endpoint, not the raw
+                # caller-supplied string. When the URL matches a registered
+                # endpoint visible to the caller, use that row's own normalized
+                # base URL (the same value owner scoping + endpoint validation
+                # already vetted) so the session dials exactly where the stored
+                # config points. The raw `endpoint` only survives for callers
+                # allowed to pass one — admins / single-user mode, where
+                # `_reject_raw_endpoint_url_for_non_admin` is a no-op and `ep`
+                # is None. Mirrors the registered-endpoint path in session_routes.
+                session_endpoint_url = (
+                    build_chat_url(normalize_base(ep.base_url)) if ep is not None else endpoint
+                )
+                # Headers come only from a matched endpoint's key; None when
+                # `ep` is None (raw admin URL or no match), so a comparison can
+                # never inherit another user's key/headers.
+                headers = build_headers(ep.api_key, ep.base_url) if (ep and ep.api_key) else None
+                resolved.append((sid, model, session_endpoint_url, headers))
+        finally:
+            db.close()
+
+        # Both endpoints validated — only now create the ephemeral [CMP]
+        # sessions and copy any resolved headers.
+        for sid, model, session_endpoint_url, headers in resolved:
             name = f"[CMP] {slot_name[sid]}" if blind else f"[CMP] {model.split('/')[-1]}"
             session_manager.create_session(
                 session_id=sid,
                 name=name,
-                endpoint_url=endpoint,
+                endpoint_url=session_endpoint_url,
                 model=model,
                 rag=False,
                 owner=user,
             )
-            # Copy API key from endpoint config
-            db = SessionLocal()
-            try:
-                from src.endpoint_resolver import build_headers, normalize_base
-                # Find matching endpoint by URL, scoped to the caller so a
-                # comparison can't borrow another user's private endpoint key.
-                base = normalize_base(endpoint)
-                ep = _owned_endpoint_by_url(db, base, user)
-                if ep and ep.api_key:
-                    s = session_manager.sessions.get(sid)
-                    if s:
-                        s.headers = build_headers(ep.api_key, ep.base_url)
-            finally:
-                db.close()
+            if headers:
+                s = session_manager.sessions.get(sid)
+                if s:
+                    s.headers = headers
 
         # Store comparison record
         db = SessionLocal()
@@ -121,8 +205,12 @@ def setup_compare_routes(session_manager: SessionManager):
                 prompt=prompt,
                 model_a=model_a,
                 model_b=model_b,
-                endpoint_a=endpoint_a,
-                endpoint_b=endpoint_b,
+                # Record the URL the session actually dials. For URL callers this
+                # is their raw input; for id-only callers (empty endpoint_a/_b)
+                # fall back to the resolved endpoint URL so the column stays
+                # meaningful and non-null. resolved is in [a, b] order.
+                endpoint_a=endpoint_a or resolved[0][2],
+                endpoint_b=endpoint_b or resolved[1][2],
                 is_blind=blind,
                 blind_mapping=json.dumps(mapping),
                 owner=user,

routes/contacts_routes.py

@@ -11,20 +11,24 @@ import uuid
 import json
 import csv
 import io
+import os
 import httpx
 from pathlib import Path
 from datetime import datetime
-from fastapi import APIRouter, Query, Depends, Response
+from urllib.parse import urljoin, urlparse, urlunparse
+
+from fastapi import APIRouter, Query, Depends, Response, HTTPException
 from typing import List, Dict, Optional
 
-from src.auth_helpers import require_user
 from core.middleware import require_admin
+from src.url_safety import check_outbound_url
 
 logger = logging.getLogger(__name__)
 
-DATA_DIR = Path(__file__).resolve().parent.parent / "data"
-SETTINGS_FILE = DATA_DIR / "settings.json"
-LOCAL_CONTACTS_FILE = DATA_DIR / "contacts.json"
+from src.constants import DATA_DIR as _DATA_DIR, SETTINGS_FILE as _SETTINGS_FILE, CONTACTS_FILE as _CONTACTS_FILE
+DATA_DIR = Path(_DATA_DIR)
+SETTINGS_FILE = Path(_SETTINGS_FILE)
+LOCAL_CONTACTS_FILE = Path(_CONTACTS_FILE)
 
 
 def _load_settings():
@@ -53,6 +57,21 @@ def _carddav_configured(cfg: Optional[Dict] = None) -> bool:
     return bool((cfg.get("url") or "").strip())
 
 
+def _validate_carddav_url(url: str) -> str:
+    cleaned = (url if isinstance(url, str) else "").strip().rstrip("/")
+    ok, reason = check_outbound_url(
+        cleaned,
+        block_private=os.getenv("CARDDAV_BLOCK_PRIVATE_IPS", "false").lower() == "true",
+    )
+    if not ok:
+        raise ValueError(f"Rejected CardDAV URL: {reason}")
+    return cleaned
+
+
+def _carddav_base_url(cfg: Dict) -> str:
+    return _validate_carddav_url(cfg.get("url") or "")
+
+
 def _normalize_contact(contact: Dict) -> Dict:
     emails = []
     for e in contact.get("emails") or ([] if not contact.get("email") else [contact.get("email")]):
@@ -219,14 +238,18 @@ _contact_cache = {"contacts": [], "fetched_at": None}
 def _abs_url(href: str) -> str:
     """Combine a multistatus <href> (an absolute path like
     /user/contacts/x.vcf) with the configured CardDAV server origin so we
-    get a fully-qualified URL to PUT/DELETE. If href is already absolute
-    (http...), return it as-is."""
-    from urllib.parse import urlparse, urlunparse
-    if href.startswith("http://") or href.startswith("https://"):
-        return href
+    get a fully-qualified URL to PUT/DELETE. Absolute hrefs are accepted only
+    for the configured origin; a cross-origin href is treated as a path on the
+    configured server so a malicious CardDAV response cannot redirect later
+    writes/deletes to cloud metadata or another host."""
     cfg = _get_carddav_config()
-    p = urlparse(cfg["url"])
-    return urlunparse((p.scheme, p.netloc, href, "", "", ""))
+    base = _carddav_base_url(cfg)
+    base_p = urlparse(base)
+    joined = urljoin(base.rstrip("/") + "/", href or "")
+    joined_p = urlparse(joined)
+    if (joined_p.scheme, joined_p.netloc) != (base_p.scheme, base_p.netloc):
+        joined = urlunparse((base_p.scheme, base_p.netloc, joined_p.path or "/", "", joined_p.query, ""))
+    return _validate_carddav_url(joined)
 
 
 # CardDAV REPORT body — pull every card's etag + raw vCard in ONE request,
@@ -297,6 +320,7 @@ def _fetch_contacts(force=False):
         return contacts
 
     try:
+        cfg["url"] = _carddav_base_url(cfg)
         auth = None
         if cfg["username"]:
             auth = (cfg["username"], cfg["password"])
@@ -353,8 +377,8 @@ def _create_contact(name: str, email: str) -> bool:
 
     contact_uid = str(uuid.uuid4())
     vcard = _build_vcard(name, email, contact_uid)
-    url = cfg["url"].rstrip("/") + "/" + contact_uid + ".vcf"
     try:
+        url = _carddav_base_url(cfg) + "/" + contact_uid + ".vcf"
         auth = None
         if cfg["username"]:
             auth = (cfg["username"], cfg["password"])
@@ -382,7 +406,7 @@ def _vcard_url(uid: str) -> str:
     escape the collection and target an arbitrary CardDAV resource."""
     from urllib.parse import quote
     cfg = _get_carddav_config()
-    return cfg["url"].rstrip("/") + "/" + quote(uid, safe="") + ".vcf"
+    return _carddav_base_url(cfg) + "/" + quote(uid, safe="") + ".vcf"
 
 
 def _import_vcards(text: str) -> Dict:
@@ -413,6 +437,11 @@ def _import_vcards(text: str) -> Dict:
         if imported:
             _save_local_contacts(contacts)
         return {"imported": imported, "failed": 0, "total": len(parsed)}
+    try:
+        base_url = _carddav_base_url(cfg)
+    except ValueError as e:
+        logger.warning("CardDAV import URL rejected: %s", e)
+        return {"imported": 0, "failed": 0, "total": 0, "error": str(e)}
     auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
     # Split into individual cards. re.split drops the BEGIN line, so we
     # re-add it. Normalize CRLF.
@@ -441,7 +470,7 @@ def _import_vcards(text: str) -> Dict:
         elif not re.search(r"^VERSION:", block, re.MULTILINE):
             block = block.replace("BEGIN:VCARD", "BEGIN:VCARD\nVERSION:4.0", 1)
         vcard = block.replace("\n", "\r\n") + "\r\n"
-        url = cfg["url"].rstrip("/") + "/" + quote(uid, safe="") + ".vcf"
+        url = base_url + "/" + quote(uid, safe="") + ".vcf"
         try:
             r = httpx.put(
                 url, data=vcard.encode("utf-8"),
@@ -601,8 +630,8 @@ def _update_contact(uid: str, name: str, emails: List[str], phones: List[str]) -
     vcard = _build_vcard(name, "", uid=uid, emails=emails, phones=phones)
     # Use the real resource href (handles externally-created contacts whose
     # filename != UID); falls back to the <uid>.vcf guess.
-    url = _resolve_resource_url(uid)
     try:
+        url = _resolve_resource_url(uid)
         auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
         r = httpx.put(
             url,
@@ -630,8 +659,8 @@ def _delete_contact(uid: str) -> bool:
         _save_local_contacts(remaining)
         return True
 
-    url = _resolve_resource_url(uid)
     try:
+        url = _resolve_resource_url(uid)
         auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
         r = httpx.delete(url, auth=auth, timeout=10)
         if r.status_code in (200, 204):
@@ -700,8 +729,11 @@ def setup_contacts_routes():
     @router.post("/import")
     async def import_vcf(data: dict, _admin: str = Depends(require_admin)):
         """Import contacts from .vcf or CSV. Body: {"vcf": "..."} or {"csv": "..."}."""
-        text = data.get("vcf") or data.get("text") or ""
-        csv_text = data.get("csv") or ""
+        # Coerce defensively: a non-string vcf/text/csv (e.g. a number or list
+        # in the JSON body) would otherwise reach .strip() and 500 with an
+        # AttributeError instead of degrading to a clean "no data" response.
+        text = str(data.get("vcf") or data.get("text") or "")
+        csv_text = str(data.get("csv") or "")
         if text.strip():
             if "BEGIN:VCARD" not in text.upper():
                 return {"success": False, "error": "No vCard data found"}
@@ -747,7 +779,13 @@ def setup_contacts_routes():
         settings = _load_settings()
         for key in ("carddav_url", "carddav_username", "carddav_password"):
             if key in data:
-                settings[key] = data[key]
+                if key == "carddav_url" and str(data[key] or "").strip():
+                    try:
+                        settings[key] = _validate_carddav_url(data[key])
+                    except ValueError as e:
+                        raise HTTPException(400, str(e))
+                else:
+                    settings[key] = data[key]
         _save_settings(settings)
         # Force re-fetch
         _contact_cache["fetched_at"] = None

routes/cookbook_helpers.py

@@ -1,16 +1,21 @@
 """cookbook_helpers.py — validators + small helpers shared by the cookbook routes.
 Extracted from cookbook_routes.py; the routes module imports the symbols it needs."""
 
+import json
 import logging
 import ntpath
 import os
 import posixpath
 import re
 import shlex
+from pathlib import Path
 
 from fastapi import HTTPException
 from pydantic import BaseModel
 
+from routes._validators import validate_remote_host, validate_ssh_port
+from core.platform_compat import _ssh_exec_argv
+
 logger = logging.getLogger(__name__)
 
 
@@ -28,20 +33,24 @@ _LOCAL_MODEL_ID_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._-]*$")
 _OLLAMA_MODEL_ID_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:/-]{0,200}$")
 # Include pattern is a glob: allow typical safe glyphs only.
 _INCLUDE_RE = re.compile(r"^[A-Za-z0-9._\-*?/\[\]]+$")
-# Remote host: user@host (optionally with :port-free hostname parts).
-_REMOTE_HOST_RE = re.compile(r"^[A-Za-z0-9._-]+@[A-Za-z0-9._-]+$")
 # HF tokens and API tokens are url-safe base64-like.
 _TOKEN_RE = re.compile(r"^[A-Za-z0-9._~+/=-]+$")
 # Session IDs we mint look like "cookbook-deadbeef" or "serve-deadbeef".
 # Anything beyond plain alphanumerics + dash + underscore could break out
 # of the shell/PowerShell contexts the value lands in.
 _SESSION_ID_RE = re.compile(r"^[A-Za-z0-9_-]{1,64}$")
-_SSH_PORT_RE = re.compile(r"^\d{1,5}$")
 _GPU_LIST_RE = re.compile(r"^\d+(?:,\d+)*$")
 # A download target directory. Absolute or ~-relative path; safe path glyphs
-# only (no quotes, shell metacharacters, or spaces) since it lands in a shell
-# command. A leading ~ is expanded to $HOME at command-build time.
-_LOCAL_DIR_RE = re.compile(r"^~?/[A-Za-z0-9._/-]*$|^~$")
+# only (no quotes or shell metacharacters). Spaces are allowed because command
+# builders pass the value through quoted shell/Python contexts. The character
+# class uses ``\w`` — Unicode word characters under Python 3's default str
+# matching — so non-ASCII folder names pass validation too: Cyrillic, accented
+# Latin, CJK, e.g. ``/Volumes/Модели`` or ``D:\AI Models\Модели``. This stays
+# shell-safe: none of ``; & | ` $ '' "" () {}`` newlines etc. are in ``[\w. -]``,
+# so injection vectors remain rejected. A leading ~ is expanded to $HOME at
+# command-build time. (Drive letters stay ASCII: ``[A-Za-z]:``.)
+_LOCAL_DIR_RE = re.compile(r"^~?(?:/[\w. -]*)+$|^~$")
+_WINDOWS_LOCAL_DIR_RE = re.compile(r"^[A-Za-z]:[\\/](?:[\w. -]+(?:[\\/][\w. -]+)*[\\/]?)?$")
 _WINDOWS_DRIVE_PATH_RE = re.compile(r"^[A-Za-z]:[\\/]")
 
 
@@ -75,14 +84,6 @@ def _validate_include(v: str | None) -> str | None:
     return v
 
 
-def _validate_remote_host(v: str | None) -> str | None:
-    if v is None or v == "":
-        return None
-    if not _REMOTE_HOST_RE.match(v):
-        raise HTTPException(400, "Invalid remote_host — must be user@host, no SSH option syntax")
-    return v
-
-
 def _validate_token(v: str | None) -> str | None:
     if v is None or v == "":
         return None
@@ -91,26 +92,43 @@ def _validate_token(v: str | None) -> str | None:
     return v
 
 
+def load_stored_hf_token(*, state_path: Path | str | None = None) -> str:
+    """Return the decrypted HF token from cookbook_state.json, else env fallback."""
+    path = Path(state_path) if state_path else Path(os.environ.get("DATA_DIR", "data")) / "cookbook_state.json"
+    token = ""
+    if path.exists():
+        try:
+            state = json.loads(path.read_text(encoding="utf-8"))
+            env = state.get("env") if isinstance(state, dict) else {}
+            if isinstance(env, dict) and env.get("hfToken"):
+                from src.secret_storage import decrypt
+                token = decrypt(env.get("hfToken") or "")
+        except Exception:
+            token = ""
+    if not token:
+        token = (os.environ.get("HF_TOKEN") or os.environ.get("HUGGING_FACE_HUB_TOKEN") or "").strip()
+    return token
+
+
 def _validate_local_dir(v: str | None) -> str | None:
     if v is None or v == "":
         return None
+    if len(v) >= 2 and v[0] == v[-1] and v[0] in {"'", '"'}:
+        v = v[1:-1]
     v = v.rstrip("/") or "/"
-    if not _LOCAL_DIR_RE.match(v):
-        raise HTTPException(400, "Invalid local_dir — must be an absolute or ~ path with no spaces or shell metacharacters")
+    if not (_LOCAL_DIR_RE.match(v) or _WINDOWS_LOCAL_DIR_RE.match(v)):
+        raise HTTPException(400, "Invalid local_dir — must be an absolute or ~ path with no shell metacharacters")
+    # Reject path segments that start with '-' (option injection). '-' is in the
+    # allowlist, so a dir like ``/models/-rf`` or ``D:\models\-rf`` could be read
+    # as a CLI flag by hf/etc. — and quoting does NOT stop a value from being
+    # parsed as an option. This is the one residual that command-build-time
+    # quoting can't cover, so the guard lives here, keeping the safety wholly
+    # inside the validator rather than relying on consumers.
+    if any(seg.startswith("-") for seg in re.split(r"[\\/]", v) if seg):
+        raise HTTPException(400, "Invalid local_dir — path segments cannot start with '-'")
     return v
 
 
-def _validate_ssh_port(v: str | None) -> str | None:
-    if v is None or v == "":
-        return None
-    if not _SSH_PORT_RE.fullmatch(str(v)):
-        raise HTTPException(400, "Invalid ssh_port")
-    port = int(v)
-    if port < 1 or port > 65535:
-        raise HTTPException(400, "Invalid ssh_port")
-    return str(port)
-
-
 def _validate_gpus(v: str | None) -> str | None:
     if v is None or v == "":
         return None
@@ -122,7 +140,7 @@ def _validate_gpus(v: str | None) -> str | None:
 def _shell_path(p: str) -> str:
     """Render a validated path for a double-quoted shell context, expanding a
     leading ~ to $HOME (single quotes wouldn't expand it). Safe because
-    _validate_local_dir already restricts the charset."""
+    _validate_local_dir already rejects quotes and shell metacharacters."""
     if p == "~":
         return '"$HOME"'
     if p.startswith("~/"):
@@ -195,6 +213,20 @@ def _pip_install_attempt(pip_cmd: str) -> str:
     )
 
 
+def _pip_command(python_cmd: str) -> str:
+    """Return a pip command for either a pip executable or a Python executable."""
+    cmd = python_cmd.strip()
+    if " -m pip" in cmd or cmd in {"pip", "pip3"}:
+        return python_cmd
+    if cmd in {"python", "python3", "python.exe"} or cmd.endswith(("/python", "/python3", "\\python.exe")):
+        return f"{python_cmd} -m pip"
+    return python_cmd
+
+
+def _pip_break_system_packages_check(pip_cmd: str) -> str:
+    return f"{pip_cmd} install --help 2>/dev/null | grep -q -- --break-system-packages"
+
+
 def _pip_install_fallback_chain(package: str, *, python_cmd: str = "python3 -m pip", upgrade: bool = False) -> str:
     """Build a bash pip install fallback chain that surfaces errors.
 
@@ -206,33 +238,44 @@ def _pip_install_fallback_chain(package: str, *, python_cmd: str = "python3 -m p
     exit code is preserved (no ``| tail`` masking) and the last 5 lines of
     pip output appear in the Cookbook log on failure.
     """
+    from core.platform_compat import IS_WINDOWS
     upgrade_flag = " -U" if upgrade else ""
     # Shell-quote the package spec: an extras spec like ``llama-cpp-python[server]``
     # contains brackets that bash would treat as a glob, so it must be quoted
     # before being embedded in the install command. Plain names (e.g.
     # ``huggingface_hub``) are returned unchanged by ``shlex.quote``.
     pkg = shlex.quote(package)
-    base = _pip_install_attempt(f"{python_cmd} install -q{upgrade_flag} {pkg}")
-    user = _pip_install_attempt(f"{python_cmd} install --user --break-system-packages -q{upgrade_flag} {pkg}")
+    # llama-cpp-python source builds are brittle on older distro pip/packaging
+    # stacks (common on WSL images). Prefer the prebuilt wheel index whenever
+    # this package is requested so dependency-install tasks are reliable.
+    if "llama-cpp-python" in package:
+        pkg += " --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu"
+
+    pip_cmd = _pip_command(python_cmd)
+    base = _pip_install_attempt(f"{pip_cmd} install -q{upgrade_flag} {pkg}")
+    user = _pip_install_attempt(f"{pip_cmd} install --user -q{upgrade_flag} {pkg}")
+    user_break_system = _pip_install_attempt(f"{pip_cmd} install --user --break-system-packages -q{upgrade_flag} {pkg}")
+    user_fallback = f"( {user} || {{ {_pip_break_system_packages_check(pip_cmd)} && {user_break_system}; }} )"
     # Derive the python executable for the venv detection check.
     # Must use the same interpreter that pip belongs to; hardcoding
     # python3 breaks when pip lives in a venv that only has "python".
-    if " -m pip" in python_cmd:
-        python_exe = python_cmd.replace(" -m pip", "")
-    elif python_cmd.strip() == "pip":
+    if " -m pip" in pip_cmd:
+        python_exe = pip_cmd.replace(" -m pip", "")
+    elif pip_cmd.strip() == "pip":
         python_exe = "python"
-    elif python_cmd.strip() == "pip3":
+    elif pip_cmd.strip() == "pip3":
         python_exe = "python3"
     else:
         python_exe = "python3"
     venv_check = f'{python_exe} -c "import sys; sys.exit(0 if sys.prefix != sys.base_prefix else 1)"'
-    # Negated: `! venv_check` succeeds (exit 0) when NOT in a venv → `&&` tries
-    # --user.  When IN a venv `! venv_check` fails → `&&` skips --user and the
+    # Negated: `! venv_check` succeeds (exit 0) when NOT in a venv -> `&&` tries
+    # --user. When IN a venv `! venv_check` fails -> `&&` skips --user and the
     # group exits non-zero, propagating the base-install failure instead of
     # masking it as success (the `|| { venv_check || … }` shape from #903
     # swallowed the exit code because venv_check's exit-0 became the group's
-    # result).
-    return f"{base} || {{ ! {venv_check} && {user}; }}"
+    # result). `--break-system-packages` is only attempted when the active pip
+    # supports it; older pip versions abort with "no such option" otherwise.
+    return f"{base} || {{ ! {venv_check} && {user_fallback}; }}"
 
 
 def _venv_safe_local_pip_install_cmd(cmd: str, *, local: bool, in_venv: bool) -> str:
@@ -263,6 +306,55 @@ def _venv_safe_local_pip_install_cmd(cmd: str, *, local: bool, in_venv: bool) ->
     return shlex.join(stripped)
 
 
+def _pip_install_command_without_break_system_packages(cmd: str) -> str:
+    try:
+        parts = shlex.split(cmd)
+    except ValueError:
+        return cmd
+    stripped = [part for part in parts if part != "--break-system-packages"]
+    return shlex.join(stripped)
+
+
+def _pip_install_help_check_from_cmd(cmd: str) -> str | None:
+    try:
+        parts = shlex.split(cmd)
+    except ValueError:
+        return None
+    try:
+        install_index = parts.index("install")
+    except ValueError:
+        return None
+    if install_index <= 0:
+        return None
+    pip_prefix = parts[:install_index]
+    return f"{shlex.join(pip_prefix + ['install', '--help'])} 2>/dev/null | grep -q -- --break-system-packages"
+
+
+def _append_pip_install_runner_lines(runner_lines: list[str], cmd: str) -> None:
+    """Append a pip install command, guarding --break-system-packages support.
+
+    The Dependencies UI may submit ``python3 -m pip install --user
+    --break-system-packages ...`` for non-venv installs. That flag is useful on
+    PEP-668-locked distros, but older pip (including Ubuntu 22.04's apt pip in
+    the NVIDIA CUDA base image) aborts with "no such option". Branch at runner
+    time so stale browser JS and remote targets are handled by the server too.
+    """
+    if "--break-system-packages" not in (cmd or ""):
+        runner_lines.append(cmd)
+        return
+    help_check = _pip_install_help_check_from_cmd(cmd)
+    without_break = _pip_install_command_without_break_system_packages(cmd)
+    if not help_check or without_break == cmd:
+        runner_lines.append(cmd)
+        return
+    runner_lines.append(f"if {help_check}; then")
+    runner_lines.append(f"  {cmd}")
+    runner_lines.append("else")
+    runner_lines.append('  echo "[odysseus] pip does not support --break-system-packages; installing without it."')
+    runner_lines.append(f"  {without_break}")
+    runner_lines.append("fi")
+
+
 def _user_shell_path_bootstrap() -> list[str]:
     return [
         'ODYSSEUS_USER_SHELL="${SHELL:-}"',
@@ -271,11 +363,14 @@ def _user_shell_path_bootstrap() -> list[str]:
         '  if [ -n "$ODYSSEUS_USER_PATH" ]; then export PATH="$ODYSSEUS_USER_PATH:$PATH"; fi',
         'fi',
         'command -v python3 >/dev/null 2>&1 || python3() { python "$@"; }',
+        'command -v python >/dev/null 2>&1 || python() { python3 "$@"; }',
     ]
 
 
-def _cached_model_scan_script(model_dirs: list[str] | None = None) -> str:
-    """Build the standalone Python scanner used by /api/model/cached."""
+def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache: str | None = None) -> str:
+    """Build the standalone Python scanner used by /api/model/cached.
+    Allows for an additional HuggingFace cache path to be scanned (i.e. Windows HF cache for local WSL envs.)
+    """
     lines = [
         "import json, os, re, shutil, subprocess, urllib.request",
         "models = []",
@@ -306,6 +401,7 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None) -> str:
         "    for root, dirs, fns in safe_walk(base):",
         "        for fn in sorted(fns):",
         "            if not fn.lower().endswith('.gguf'): continue",
+        "            if fn.startswith('._'): continue  # macOS AppleDouble sidecar, not a real GGUF",
         "            fp = os.path.join(root, fn)",
         "            try: size = os.path.getsize(fp)",
         "            except Exception: size = 0",
@@ -338,6 +434,15 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None) -> str:
         "                if f.is_file(): nf += 1; sz += f.stat().st_size",
         "                if f.name.endswith('.incomplete'): ic = True",
         "        snap = os.path.join(cache, d, 'snapshots')",
+        "        # Windows HF cache stores files directly in snapshots/; blobs/ may be empty.",
+        "        # Fallback: scan snapshots for real files when blobs yielded nothing.",
+        "        if sz == 0 and os.path.isdir(snap):",
+        "            for sd in os.listdir(snap):",
+        "                sf = os.path.join(snap, sd)",
+        "                if not os.path.isdir(sf): continue",
+        "                for f in os.scandir(sf):",
+        "                    if f.is_file(): nf += 1; sz += f.stat().st_size",
+        "                    if f.name.endswith('.incomplete'): ic = True",
         "        is_diffusion = False; gguf_files = []",
         "        if os.path.isdir(snap):",
         "            for sd in os.listdir(snap):",
@@ -346,6 +451,21 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None) -> str:
         "                if os.path.exists(os.path.join(sf, 'model_index.json')): is_diffusion = True",
         "                for f in collect_ggufs(sf): f['rel_path'] = sd + '/' + f['rel_path']; gguf_files.append(f)",
         "        models.append({'repo_id':rid,'size_bytes':sz,'nb_files':nf,'has_incomplete':ic,'path':cache,'is_diffusion':is_diffusion,'is_gguf':bool(gguf_files),'gguf_files':gguf_files})",
+        "def hf_cache_paths():",
+        "    candidates = []",
+        "    def add(p):",
+        "        if not p: return",
+        "        p = os.path.expanduser(p)",
+        "        if p not in candidates: candidates.append(p)",
+        "    add(os.environ.get('HUGGINGFACE_HUB_CACHE'))",
+        "    hf_home = os.environ.get('HF_HOME')",
+        "    if hf_home: add(os.path.join(hf_home, 'hub'))",
+        "    add('~/.cache/huggingface/hub')",
+        "    # Docker images mount ./data/huggingface at /app/.cache/huggingface.",
+        "    # When HOME is /root, expanduser() misses that persisted cache.",
+        "    add('/app/.cache/huggingface/hub')",
+        f"    add({add_hf_cache!r})" if add_hf_cache else "",
+        "    return candidates",
         "def scan_dir(p):",
         "    if not os.path.isdir(p) or not safe_path(p): return",
         "    for d in sorted(os.listdir(p)):",
@@ -409,7 +529,7 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None) -> str:
         "            seen.add(name)",
         "            models.append({'repo_id':name,'size_bytes':size_bytes,'nb_files':1,'has_incomplete':False,'path':'ollama','backend':'ollama','is_ollama':True})",
         "        return",
-        "scan_hf(os.path.expanduser('~/.cache/huggingface/hub'))",
+        "for _hf_cache in hf_cache_paths(): scan_hf(_hf_cache)",
         "scan_ollama()",
         "scan_ollama_api()",
     ]
@@ -525,6 +645,7 @@ def _validate_serve_cmd(v: str | None) -> str | None:
     # Backticks and raw newlines are never legitimate here.
     if any(c in v for c in ("`", "\n", "\r")):
         raise HTTPException(400, "Invalid characters in cmd")
+
     # Known GGUF launcher prelude → validate the serve invocation(s) it guards.
     m = _GGUF_PRELUDE_RE.match(v)
     if m:
@@ -533,9 +654,19 @@ def _validate_serve_cmd(v: str | None) -> str | None:
         for part in rest.split("||"):
             _check_serve_binary(part.strip())
         return v
+
     # Otherwise: a single invocation — no shell metacharacters allowed.
+    # Temporarily replace safe $(printf %s ...) expressions with a placeholder
+    # to avoid triggering the metacharacter/command-injection checks.
+    cleaned_v = v
+    printf_matches = list(re.finditer(r"\$\(\s*printf\s+%s\s+([^\n()]*?)\)", v))
+    for match in printf_matches:
+        inner = match.group(1)
+        if not any(c in inner for c in (";", "&&", "||", "$(", "`")):
+            cleaned_v = cleaned_v.replace(match.group(0), "/placeholder/safe/path.gguf")
+
     # (`$(` was the original intent; bare `$` is fine for shell-safe paths.)
-    if any(c in v for c in (";", "&&", "||", "$(")):
+    if any(c in cleaned_v for c in (";", "&&", "||", "$(")):
         raise HTTPException(400, "Invalid characters in cmd")
     _check_serve_binary(v)
     return v
@@ -546,12 +677,34 @@ def _append_serve_preflight_exit_lines(runner_lines: list[str], *, keep_shell_op
     runner_lines.append('if [ -n "$ODYSSEUS_PREFLIGHT_EXIT" ]; then')
     runner_lines.append('  echo ""; echo "=== Process exited with code $ODYSSEUS_PREFLIGHT_EXIT ==="')
     if keep_shell_open:
+        # Decouple the post-crash interactive shell from the persistent log
+        # file. fds 3/4 were saved BEFORE the tee redirect at the top of
+        # the runner; restoring them here means the neofetch banner the
+        # user's .zshrc prints lands on the tmux pane only, not in the
+        # log file the agent's tail_serve_output reads.
+        runner_lines.append('  exec 1>&3 2>&4 3>&- 4>&- 2>/dev/null || true')
+        runner_lines.append('  sleep 0.2  # let tee child flush + exit')
         runner_lines.append('  exec "${SHELL:-/bin/bash}"')
     else:
         runner_lines.append('  exit "$ODYSSEUS_PREFLIGHT_EXIT"')
     runner_lines.append('fi')
 
 
+def _append_vllm_linux_preflight_lines(runner_lines: list[str]) -> None:
+    """Append Linux vLLM readiness lines that identify the runtime being used."""
+    # Keep the user install bin visible for Odysseus-managed `pip install --user`
+    # installs, but then report the actual CLI path so external runtimes are clear.
+    runner_lines.append('export PATH="$HOME/.local/bin:$PATH"')
+    runner_lines.append('ODYSSEUS_VLLM_BIN="$(command -v vllm 2>/dev/null || true)"')
+    runner_lines.append('if [ -z "$ODYSSEUS_VLLM_BIN" ]; then')
+    runner_lines.append('  echo "ERROR: vLLM is not installed."')
+    runner_lines.append('  ODYSSEUS_PREFLIGHT_EXIT=127')
+    runner_lines.append('else')
+    runner_lines.append('  echo "[odysseus] vLLM CLI: $ODYSSEUS_VLLM_BIN"')
+    runner_lines.append('  ODYSSEUS_VLLM_VERSION="$("$ODYSSEUS_VLLM_BIN" --version 2>&1 | head -n 1 || true)"')
+    runner_lines.append('  if [ -n "$ODYSSEUS_VLLM_VERSION" ]; then echo "[odysseus] vLLM version: $ODYSSEUS_VLLM_VERSION"; fi')
+    runner_lines.append('fi')
+
 def _append_serve_exit_code_lines(
     runner_lines: list[str],
     *,
@@ -563,7 +716,11 @@ def _append_serve_exit_code_lines(
     if is_pip_install:
         runner_lines.append('if [ $ODYSSEUS_CMD_EXIT -eq 0 ]; then echo ""; echo "DOWNLOAD_OK"; fi')
     if keep_shell_open:
-        runner_lines.append('echo ""; echo "=== Process exited with code $ODYSSEUS_CMD_EXIT ==="; exec "${SHELL:-/bin/bash}"')
+        runner_lines.append('echo ""; echo "=== Process exited with code $ODYSSEUS_CMD_EXIT ==="')
+        # See preflight branch above for the rationale on restoring fds 3/4.
+        runner_lines.append('exec 1>&3 2>&4 3>&- 4>&- 2>/dev/null || true')
+        runner_lines.append('sleep 0.2  # let tee child flush + exit')
+        runner_lines.append('exec "${SHELL:-/bin/bash}"')
     else:
         runner_lines.append('echo ""; echo "=== Process exited with code $ODYSSEUS_CMD_EXIT ==="')
         runner_lines.append('exit "$ODYSSEUS_CMD_EXIT"')
@@ -647,6 +804,7 @@ def _llama_cpp_rebuild_cmd() -> str:
 
 class ModelDownloadRequest(BaseModel):
     repo_id: str
+    backend: str | None = None  # "hf" (default) or "ollama"
     include: str | None = None  # glob pattern e.g. "*Q4_K_M*"
     hf_token: str | None = None
     env_prefix: str | None = None  # e.g. "source ~/venv/bin/activate"
@@ -793,3 +951,172 @@ def _ssh_ps(host, script_path, port=None):
 
 # Windows session dir — stored in user's temp on the remote
 WIN_SESSION_DIR = "$env:TEMP\\\\odysseus-sessions"
+
+
+def _diagnose_serve_output(text: str) -> dict | None:
+    """Server-side mirror of the Cookbook UI's common serve diagnoses.
+
+    The browser uses cookbook-diagnosis.js for clickable fixes. This gives
+    the agent/tool path the same structured signal so it can retry with an
+    adjusted command instead of guessing from raw tmux output.
+    """
+    if not text:
+        return None
+    tail = text[-6000:]
+    patterns = [
+        (
+            r"No available memory for the cache blocks|Available KV cache memory:.*-",
+            "No GPU memory left for KV cache after loading model.",
+            [
+                {"label": "retry with GPU memory utilization 0.95", "op": "replace", "flag": "--gpu-memory-utilization", "value": "0.95"},
+                {"label": "retry with context 2048", "op": "replace", "flag": "--max-model-len", "value": "2048"},
+            ],
+        ),
+        (
+            r"CUDA out of memory|torch\.cuda\.OutOfMemoryError|CUDA error: out of memory|warming up sampler|max_num_seqs.*gpu_memory_utilization",
+            "GPU ran out of memory during startup or warmup.",
+            [
+                {"label": "retry with context 4096", "op": "replace", "flag": "--max-model-len", "value": "4096"},
+                {"label": "retry with GPU memory utilization 0.80", "op": "replace", "flag": "--gpu-memory-utilization", "value": "0.80"},
+                {"label": "retry with --enforce-eager", "op": "append", "arg": "--enforce-eager"},
+            ],
+        ),
+        (
+            r"not divisib|must be divisible|attention heads.*divisible",
+            "Tensor parallel size is incompatible with the model.",
+            [
+                {"label": "retry with tensor parallel size 1", "op": "replace", "flag": "--tensor-parallel-size", "value": "1"},
+                {"label": "retry with tensor parallel size 2", "op": "replace", "flag": "--tensor-parallel-size", "value": "2"},
+            ],
+        ),
+        (
+            r"KV cache.*too (small|large)|max_model_len.*exceeds|maximum.*context",
+            "Context length is too large for available GPU memory.",
+            [
+                {"label": "retry with context 8192", "op": "replace", "flag": "--max-model-len", "value": "8192"},
+                {"label": "retry with context 4096", "op": "replace", "flag": "--max-model-len", "value": "4096"},
+            ],
+        ),
+        (
+            r"enable-auto-tool-choice requires --tool-call-parser",
+            "Auto tool choice requires an explicit tool call parser.",
+            [{"label": "retry with Hermes tool parser", "op": "append", "arg": "--tool-call-parser hermes"}],
+        ),
+        (
+            r"Please pass.*trust.remote.code=True|contains custom code which must be executed to correctly load|does not recognize this architecture|model type.*but Transformers does not",
+            "Model requires custom code or newer model support.",
+            [{"label": "retry with --trust-remote-code", "op": "append", "arg": "--trust-remote-code"}],
+        ),
+        (
+            r"There is no module or parameter named ['\"]lm_head\.input_scale['\"]|lm_head\.input_scale|weight_scale_2",
+            "vLLM cannot load this ModelOpt LM-head quantized checkpoint with the current runtime.",
+            [
+                {
+                    "label": "upgrade vLLM through the environment that provides this CLI, or use a compatible checkpoint",
+                    "op": "manual",
+                }
+            ],
+        ),
+        (
+            r"Either a revision or a version must be specified|transformers\.integrations\.hub_kernels|kernels/layer",
+            "vLLM/Transformers kernel package mismatch.",
+            [{"label": "update vLLM, Transformers, and kernels on this server", "op": "dependency", "package": "vllm transformers kernels"}],
+        ),
+        (
+            r"Address already in use|bind.*address.*in use",
+            "Port is already in use.",
+            [{"label": "retry on port 8001", "op": "replace", "flag": "--port", "value": "8001"}],
+        ),
+        (
+            r"No CUDA GPUs are available|no GPU.*found|CUDA_VISIBLE_DEVICES.*invalid",
+            "No GPUs are visible to the serve process.",
+            [{"label": "clear Cookbook GPU selection or choose available GPUs", "op": "settings", "field": "gpus", "value": ""}],
+        ),
+        (
+            r"Failed to infer device type|NVML Shared Library Not Found|No module named 'amdsmi'|platform is not available",
+            "vLLM could not find a supported GPU (CUDA or ROCm). "
+            "This machine may have integrated or unsupported graphics only.",
+            [
+                {"label": "switch to llama.cpp (CPU/Metal, works without a discrete GPU)", "op": "manual"},
+                {"label": "switch to Ollama (CPU/Metal, works without a discrete GPU)", "op": "manual"},
+            ],
+        ),
+        (
+            r"vllm.*command not found|No module named vllm|ERROR: vLLM is not installed",
+            "vLLM is not installed or not in PATH on this server.",
+            [{"label": "install vLLM in Cookbook Dependencies", "op": "dependency", "package": "vllm"}],
+        ),
+        (
+            r"sglang.*command not found|No module named sglang|SGLang is not installed",
+            "SGLang is not installed or not in PATH on this server.",
+            [{"label": "install SGLang in Cookbook Dependencies", "op": "dependency", "package": "sglang[all]"}],
+        ),
+        (
+            r"llama-server.*command not found|llama\.cpp.*not found|No module named.*llama_cpp|No module named 'starlette_context'|git: command not found|cmake: command not found",
+            "llama.cpp / llama-cpp-python dependencies are missing.",
+            [{"label": "install llama.cpp dependencies or llama-cpp-python[server]", "op": "dependency", "package": "llama-cpp-python[server]"}],
+        ),
+        (
+            r"No GGUF found on this host|no \.gguf file|No GGUF file found",
+            "No GGUF file found for this model on this host. The llama.cpp backend needs a .gguf file.",
+            [{"label": "download a GGUF build of this model (repo name usually ends in -GGUF, file like Q4_K_M.gguf)", "op": "manual"}],
+        ),
+        (
+            r"No module named 'torch'|No module named torch|No module named 'diffusers'|No module named diffusers",
+            "Diffusion serving requires PyTorch and diffusers.",
+            [{"label": "install diffusers[torch] in Cookbook Dependencies", "op": "dependency", "package": "diffusers[torch]"}],
+        ),
+        (
+            r"403 Forbidden|401 Unauthorized|Access to model.*is restricted|gated repo|not in the authorized list|awaiting a review",
+            "Model access is gated or unauthorized.",
+            [{"label": "set HF token and request model access on HuggingFace", "op": "manual"}],
+        ),
+    ]
+    for pattern, message, suggestions in patterns:
+        if re.search(pattern, tail, re.I):
+            return {"message": message, "suggestions": suggestions}
+    if re.search(r"Traceback \(most recent call last\)", tail, re.I) and not re.search(
+        r"Application startup complete|GET /v1/|Uvicorn running on", tail, re.I
+    ):
+        return {
+            "message": "Python traceback detected during serve startup.",
+            "suggestions": [{"label": "inspect traceback and retry with adjusted backend/settings", "op": "manual"}],
+        }
+    return None
+
+
+async def run_ssh_command_async(
+    remote: str,
+    ssh_port: str | None,
+    remote_cmd: str,
+    *,
+    timeout: float,
+    connect_timeout: int | None = None,
+    strict_host_key_checking: bool | None = None,
+    stdin_data: bytes | None = None,
+) -> tuple[int, bytes, bytes]:
+    """Run an ssh command with centralized timeout and stderr/stdout capture.
+    Async version of core.platform_compat.run_ssh_command_sync.
+    """
+    import asyncio
+    proc = await asyncio.create_subprocess_exec(
+        *_ssh_exec_argv(
+            remote,
+            ssh_port,
+            remote_cmd=remote_cmd,
+            connect_timeout=connect_timeout,
+            strict_host_key_checking=strict_host_key_checking,
+        ),
+        stdin=asyncio.subprocess.PIPE if stdin_data is not None else None,
+        stdout=asyncio.subprocess.PIPE,
+        stderr=asyncio.subprocess.PIPE,
+    )
+    try:
+        stdout, stderr = await asyncio.wait_for(
+            proc.communicate(input=stdin_data), timeout=timeout
+        )
+    except asyncio.TimeoutError:
+        proc.kill()
+        await proc.communicate()
+        raise
+    return proc.returncode or 0, stdout, stderr

routes/cookbook_output.py

@@ -0,0 +1,19 @@
+"""Pure helpers for shaping cookbook task output for the status response.
+
+Kept dependency-free (no FastAPI / SQLAlchemy imports) so the behavior can be
+unit-tested without standing up the whole app.
+"""
+
+
+def error_aware_output_tail(full_snapshot: str, status: str) -> str:
+    """Return the trailing slice of a task log for the status response.
+
+    Failed tasks return the last 50 lines so the "Copy last 50 lines" action
+    surfaces the actual error context (stack traces, build output). Running and
+    other non-error tasks keep the cheaper 12-line tail to limit the payload on
+    the 10s polling interval.
+    """
+    if not full_snapshot:
+        return ""
+    tail_lines = 50 if status == "error" else 12
+    return "\n".join(full_snapshot.splitlines()[-tail_lines:])

routes/copilot_routes.py

@@ -20,39 +20,26 @@ All routes are admin-gated (endpoint/provider management is an admin action).
 """
 
 import json
-import time
 import uuid
 import logging
-import threading
 from typing import Dict, Optional
 
 import httpx
-from fastapi import APIRouter, Request, Form, HTTPException
+from fastapi import HTTPException, Request
 
 from core.database import SessionLocal, ModelEndpoint
-from core.middleware import require_admin
+from routes.device_flow import (
+    DeviceFlowPoll,
+    DeviceFlowStart,
+    PendingDeviceFlowStore,
+    create_device_flow_router,
+)
 from src.auth_helpers import get_current_user
 from src import copilot
 
 logger = logging.getLogger(__name__)
 
-# Pending device-flow logins, keyed by an opaque poll_id. The device_code is a
-# bearer-like secret, so it lives here (server memory) rather than in the
-# browser. Entries expire with the GitHub device code.
-#
-# NOTE: this is per-process state. The device flow assumes a single worker
-# (Odysseus' default): with multiple uvicorn workers, the poll request can land
-# on a worker that never saw the start, returning "Unknown or expired login
-# session". Move this to a shared store (DB/Redis) if running multi-worker.
-_PENDING: Dict[str, Dict] = {}
-_PENDING_LOCK = threading.Lock()
-
-
-def _prune_expired() -> None:
-    now = time.time()
-    with _PENDING_LOCK:
-        for k in [k for k, v in _PENDING.items() if v.get("expires_at", 0) < now]:
-            _PENDING.pop(k, None)
+_DEVICE_FLOW_STORE = PendingDeviceFlowStore()
 
 
 def _provision_endpoint(token: str, base: str, owner: Optional[str]) -> Dict:
@@ -112,112 +99,75 @@ def _provision_endpoint(token: str, base: str, owner: Optional[str]) -> Dict:
     return result
 
 
-def setup_copilot_routes() -> APIRouter:
-    router = APIRouter(prefix="/api/copilot", tags=["copilot"])
+def _start_device_flow(request: Request, form) -> DeviceFlowStart:
+    host = copilot.GITHUB_HOST
+    ent = str(form.get("enterprise_url") or "").strip()
+    if ent:
+        host = copilot.normalize_domain(ent)
+    try:
+        data = copilot.request_device_code(host)
+    except httpx.HTTPStatusError as e:
+        status = e.response.status_code if e.response is not None else "unknown"
+        raise HTTPException(502, f"GitHub device-code request failed (HTTP {status})")
+    except Exception as e:
+        raise HTTPException(502, f"GitHub device-code request failed: {e}")
 
-    @router.post("/device/start")
-    def device_start(request: Request, enterprise_url: str = Form("")):
-        require_admin(request)
-        _prune_expired()
-        host = copilot.GITHUB_HOST
-        ent = (enterprise_url or "").strip()
-        if ent:
-            host = copilot.normalize_domain(ent)
-        try:
-            data = copilot.request_device_code(host)
-        except httpx.HTTPStatusError as e:
-            status = e.response.status_code if e.response is not None else "unknown"
-            raise HTTPException(502, f"GitHub device-code request failed (HTTP {status})")
-        except Exception as e:
-            raise HTTPException(502, f"GitHub device-code request failed: {e}")
+    device_code = data.get("device_code")
+    if not device_code:
+        raise HTTPException(502, "GitHub did not return a device code")
 
-        device_code = data.get("device_code")
-        if not device_code:
-            raise HTTPException(502, "GitHub did not return a device code")
-        interval = int(data.get("interval") or 5)
-        expires_in = int(data.get("expires_in") or 900)
-        poll_id = uuid.uuid4().hex
-        with _PENDING_LOCK:
-            _PENDING[poll_id] = {
-                "device_code": device_code,
-                "host": host,
-                "enterprise_url": ent,
-                "interval": interval,
-                "owner": get_current_user(request) or None,
-                "expires_at": time.time() + expires_in,
-                "next_poll_at": 0.0,
-            }
-        # verification_uri_complete embeds the user code, so the browser tab we
-        # open lands the user straight on GitHub's "Authorize" screen with the
-        # code pre-filled — one click, no manual code entry.
-        return {
-            "poll_id": poll_id,
+    # verification_uri_complete embeds the user code, so the browser tab we
+    # open lands the user straight on GitHub's "Authorize" screen with the
+    # code pre-filled — one click, no manual code entry.
+    return DeviceFlowStart(
+        pending={
+            "device_code": device_code,
+            "host": host,
+            "enterprise_url": ent,
+            "owner": get_current_user(request) or None,
+        },
+        response={
             "user_code": data.get("user_code"),
             "verification_uri": data.get("verification_uri"),
             "verification_uri_complete": data.get("verification_uri_complete"),
-            "interval": interval,
-            "expires_in": expires_in,
-        }
+        },
+        interval=int(data.get("interval") or 5),
+        expires_in=int(data.get("expires_in") or 900),
+    )
 
-    @router.post("/device/poll")
-    def device_poll(request: Request, poll_id: str = Form(...)):
-        require_admin(request)
-        _prune_expired()
-        with _PENDING_LOCK:
-            pending = _PENDING.get(poll_id)
-        if not pending:
-            raise HTTPException(404, "Unknown or expired login session")
 
-        # Enforce GitHub's polling interval server-side so a chatty client
-        # can't trip slow_down.
-        now = time.time()
-        if now < pending.get("next_poll_at", 0):
-            return {"status": "pending"}
+def _poll_device_flow(_request: Request, pending: Dict) -> DeviceFlowPoll:
+    try:
+        data = copilot.poll_access_token(pending["host"], pending["device_code"])
+    except Exception as e:
+        return DeviceFlowPoll.pending(f"poll error: {e}")
 
+    token = data.get("access_token")
+    if token:
+        base = copilot.enterprise_base(pending["enterprise_url"]) if pending["enterprise_url"] else copilot.COPILOT_BASE
         try:
-            data = copilot.poll_access_token(pending["host"], pending["device_code"])
+            result = _provision_endpoint(token, base, pending["owner"])
         except Exception as e:
-            return {"status": "pending", "detail": f"poll error: {e}"}
+            logger.exception("Copilot endpoint provisioning failed")
+            raise HTTPException(500, f"Login succeeded but provisioning failed: {e}")
+        return DeviceFlowPoll.authorized(result)
 
-        token = data.get("access_token")
-        if token:
-            base = copilot.enterprise_base(pending["enterprise_url"]) if pending["enterprise_url"] else copilot.COPILOT_BASE
-            try:
-                result = _provision_endpoint(token, base, pending["owner"])
-            except Exception as e:
-                logger.exception("Copilot endpoint provisioning failed")
-                with _PENDING_LOCK:
-                    _PENDING.pop(poll_id, None)
-                raise HTTPException(500, f"Login succeeded but provisioning failed: {e}")
-            with _PENDING_LOCK:
-                _PENDING.pop(poll_id, None)
-            return {"status": "authorized", "endpoint": result}
+    err = data.get("error")
+    if err == "authorization_pending":
+        return DeviceFlowPoll.pending()
+    if err == "slow_down":
+        return DeviceFlowPoll.slow_down(int(data.get("interval") or 0) or None)
+    if err in ("expired_token", "access_denied"):
+        return DeviceFlowPoll.failed(err)
+    # Unknown error — surface but keep the session for another try.
+    return DeviceFlowPoll.pending(err or "unknown")
 
-        err = data.get("error")
-        if err == "authorization_pending":
-            with _PENDING_LOCK:
-                if poll_id in _PENDING:
-                    _PENDING[poll_id]["next_poll_at"] = now + pending["interval"]
-            return {"status": "pending"}
-        if err == "slow_down":
-            new_interval = int(data.get("interval") or (pending["interval"] + 5))
-            with _PENDING_LOCK:
-                if poll_id in _PENDING:
-                    _PENDING[poll_id]["interval"] = new_interval
-                    _PENDING[poll_id]["next_poll_at"] = now + new_interval
-            return {"status": "pending"}
-        if err in ("expired_token", "access_denied"):
-            with _PENDING_LOCK:
-                _PENDING.pop(poll_id, None)
-            return {"status": "failed", "error": err}
-        # Unknown error — surface but keep the session for another try.
-        return {"status": "pending", "detail": err or "unknown"}
 
-    @router.post("/device/cancel")
-    def device_cancel(request: Request, poll_id: str = Form(...)):
-        require_admin(request)
-        with _PENDING_LOCK:
-            _PENDING.pop(poll_id, None)
-        return {"status": "cancelled"}
-
-    return router
+def setup_copilot_routes():
+    return create_device_flow_router(
+        prefix="/api/copilot",
+        tags=["copilot"],
+        store=_DEVICE_FLOW_STORE,
+        start_flow=_start_device_flow,
+        poll_flow=_poll_device_flow,
+    )

routes/device_flow.py

@@ -0,0 +1,193 @@
+"""Shared OAuth/device-flow route scaffolding for provider setup."""
+
+from __future__ import annotations
+
+import inspect
+import threading
+import time
+import uuid
+from dataclasses import dataclass
+from typing import Any, Callable, Iterable, Mapping, Optional
+
+from fastapi import APIRouter, Form, HTTPException, Request
+
+from core.middleware import require_admin
+
+
+@dataclass(frozen=True)
+class DeviceFlowStart:
+    """Provider-specific start result consumed by the shared route wrapper."""
+
+    pending: Mapping[str, Any]
+    response: Mapping[str, Any]
+    interval: int = 5
+    expires_in: int = 900
+
+
+@dataclass(frozen=True)
+class DeviceFlowPoll:
+    """Normalized provider poll outcome."""
+
+    status: str
+    endpoint: Optional[Mapping[str, Any]] = None
+    error: Optional[str] = None
+    detail: Optional[str] = None
+    interval: Optional[int] = None
+
+    @classmethod
+    def pending(cls, detail: Optional[str] = None) -> "DeviceFlowPoll":
+        return cls(status="pending", detail=detail)
+
+    @classmethod
+    def slow_down(cls, interval: Optional[int] = None, detail: Optional[str] = None) -> "DeviceFlowPoll":
+        return cls(status="slow_down", interval=interval, detail=detail)
+
+    @classmethod
+    def authorized(cls, endpoint: Mapping[str, Any]) -> "DeviceFlowPoll":
+        return cls(status="authorized", endpoint=endpoint)
+
+    @classmethod
+    def failed(cls, error: str) -> "DeviceFlowPoll":
+        return cls(status="failed", error=error)
+
+
+class PendingDeviceFlowStore:
+    """Thread-safe in-memory pending device-flow store.
+
+    Device codes and provider-side secrets stay inside this process. Each entry
+    stores provider payload separately from poll metadata so provider callbacks
+    only receive the fields they created.
+    """
+
+    def __init__(self, *, time_func: Callable[[], float] = time.time):
+        self._pending: dict[str, dict[str, Any]] = {}
+        self._lock = threading.Lock()
+        self._time = time_func
+
+    def _now(self) -> float:
+        return float(self._time())
+
+    def prune_expired(self) -> None:
+        now = self._now()
+        with self._lock:
+            for key in [k for k, v in self._pending.items() if v.get("expires_at", 0) < now]:
+                self._pending.pop(key, None)
+
+    def add(self, payload: Mapping[str, Any], *, interval: int, expires_in: int) -> str:
+        self.prune_expired()
+        poll_id = uuid.uuid4().hex
+        with self._lock:
+            self._pending[poll_id] = {
+                "payload": dict(payload),
+                "interval": max(int(interval or 5), 1),
+                "expires_at": self._now() + max(int(expires_in or 900), 1),
+                "next_poll_at": 0.0,
+            }
+        return poll_id
+
+    def get_payload(self, poll_id: str) -> Optional[dict[str, Any]]:
+        self.prune_expired()
+        with self._lock:
+            entry = self._pending.get(poll_id)
+            if entry is None:
+                return None
+            return dict(entry.get("payload") or {})
+
+    def is_throttled(self, poll_id: str) -> bool:
+        with self._lock:
+            entry = self._pending.get(poll_id)
+            return bool(entry and self._now() < float(entry.get("next_poll_at") or 0))
+
+    def schedule_next(self, poll_id: str) -> None:
+        now = self._now()
+        with self._lock:
+            entry = self._pending.get(poll_id)
+            if entry is not None:
+                entry["next_poll_at"] = now + int(entry.get("interval") or 5)
+
+    def slow_down(self, poll_id: str, interval: Optional[int] = None) -> None:
+        now = self._now()
+        with self._lock:
+            entry = self._pending.get(poll_id)
+            if entry is not None:
+                new_interval = int(interval or (int(entry.get("interval") or 5) + 5))
+                entry["interval"] = max(new_interval, 1)
+                entry["next_poll_at"] = now + entry["interval"]
+
+    def pop(self, poll_id: str) -> None:
+        with self._lock:
+            self._pending.pop(poll_id, None)
+
+
+async def _maybe_await(value: Any) -> Any:
+    if inspect.isawaitable(value):
+        return await value
+    return value
+
+
+def _pending_response(detail: Optional[str] = None) -> dict[str, Any]:
+    response: dict[str, Any] = {"status": "pending"}
+    if detail:
+        response["detail"] = detail
+    return response
+
+
+def create_device_flow_router(
+    *,
+    prefix: str,
+    tags: Iterable[str],
+    store: PendingDeviceFlowStore,
+    start_flow: Callable[[Request, Mapping[str, Any]], DeviceFlowStart],
+    poll_flow: Callable[[Request, Mapping[str, Any]], DeviceFlowPoll],
+) -> APIRouter:
+    """Create standard `/device/start|poll|cancel` routes for a provider."""
+
+    router = APIRouter(prefix=prefix, tags=list(tags))
+
+    @router.post("/device/start")
+    async def device_start(request: Request):
+        require_admin(request)
+        form = await request.form()
+        start = await _maybe_await(start_flow(request, form))
+        interval = int(start.interval or 5)
+        expires_in = int(start.expires_in or 900)
+        poll_id = store.add(start.pending, interval=interval, expires_in=expires_in)
+        response = dict(start.response)
+        response.update({"poll_id": poll_id, "interval": interval, "expires_in": expires_in})
+        return response
+
+    @router.post("/device/poll")
+    async def device_poll(request: Request, poll_id: str = Form(...)):
+        require_admin(request)
+        payload = store.get_payload(poll_id)
+        if payload is None:
+            raise HTTPException(404, "Unknown or expired login session")
+        if store.is_throttled(poll_id):
+            return {"status": "pending"}
+
+        try:
+            outcome = await _maybe_await(poll_flow(request, payload))
+        except Exception:
+            store.pop(poll_id)
+            raise
+
+        if outcome.status == "authorized":
+            store.pop(poll_id)
+            return {"status": "authorized", "endpoint": dict(outcome.endpoint or {})}
+        if outcome.status == "failed":
+            store.pop(poll_id)
+            return {"status": "failed", "error": outcome.error or "denied"}
+        if outcome.status == "slow_down":
+            store.slow_down(poll_id, outcome.interval)
+            return _pending_response(outcome.detail)
+
+        store.schedule_next(poll_id)
+        return _pending_response(outcome.detail)
+
+    @router.post("/device/cancel")
+    def device_cancel(request: Request, poll_id: str = Form(...)):
+        require_admin(request)
+        store.pop(poll_id)
+        return {"status": "cancelled"}
+
+    return router

routes/diagnostics_routes.py

@@ -16,9 +16,18 @@ def setup_diagnostics_routes(
     rag_manager,
     rag_available: bool,
     research_handler,
+    memory_vector=None,
 ) -> APIRouter:
     router = APIRouter(tags=["diagnostics"])
 
+    @router.get("/api/diagnostics/services")
+    async def get_service_health(request: Request) -> Dict[str, Any]:
+        """Consolidated degraded-state report for ChromaDB, SearXNG, email,
+        ntfy, and provider endpoints. Non-intrusive probes — safe to poll."""
+        require_admin(request)
+        from src.service_health import collect_service_health
+        return await collect_service_health(rag_manager, memory_vector)
+
     @router.get("/api/db/stats")
     async def get_database_stats(request: Request) -> Dict[str, Any]:
         require_admin(request)

routes/document_routes.py

@@ -7,14 +7,24 @@ from typing import Dict, Any, List, Optional
 
 from fastapi import APIRouter, HTTPException, Query, Request, UploadFile, File, Form
 
-from sqlalchemy import func
+from sqlalchemy import case, func, or_
 from core.database import SessionLocal, Document, DocumentVersion
 from core.database import Session as DbSession
 from src.auth_helpers import get_current_user
+from src.constants import MAIL_ATTACHMENTS_DIR
 
 logger = logging.getLogger(__name__)
 
 
+def _get_session_or_404(db, session_id: str, user: Optional[str]):
+    session = db.query(DbSession).filter(DbSession.id == session_id).first()
+    if not session:
+        raise HTTPException(404, "Session not found")
+    if user and session.owner != user:
+        raise HTTPException(404, "Session not found")
+    return session
+
+
 def _aggregate_language_facets(lang_rows):
     """Sum document counts per display language for the library facet.
 
@@ -30,6 +40,19 @@ def _aggregate_language_facets(lang_rows):
     return out
 
 
+def _library_language_for_document(doc: Document) -> str:
+    """Return the display language used by the document library.
+
+    PDF documents are stored as markdown wrappers so the editor can preserve
+    extracted text, form fields, and annotations. The library should still
+    identify them as PDFs instead of exposing that internal wrapper format.
+    """
+    from src.pdf_form_doc import find_source_upload_id
+
+    if find_source_upload_id(doc.current_content or ""):
+        return "pdf"
+    return doc.language or "text"
+
 
 from routes.document_helpers import (
     DocumentCreate, DocumentUpdate, DocumentPatch,
@@ -69,17 +92,12 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             # the doc is owner-stamped, so it lives in the library on its own.
             session = None
             if req.session_id:
-                session = db.query(DbSession).filter(DbSession.id == req.session_id).first()
-                if not session:
-                    raise HTTPException(404, "Session not found")
                 # Match the lenient ownership model the rest of the app uses
                 # (see _owner_filter): only block when an AUTHENTICATED user is
                 # writing into a DIFFERENT user's session. In single-user /
-                # unconfigured / localhost-bypass mode the middleware leaves
-                # current_user unset (None), and those sessions are already
-                # served freely everywhere else.
-                if user and session.owner and session.owner != user:
-                    raise HTTPException(403, "Cannot create document in another user's session")
+                # unconfigured / localhost-bypass mode, falsey users preserve
+                # the existing lenient path.
+                session = _get_session_or_404(db, req.session_id, user)
 
             doc_id = str(uuid.uuid4())
             ver_id = str(uuid.uuid4())
@@ -90,10 +108,10 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             # to markdown for prose.
             language = req.language
             if not language:
-                from src.tool_implementations import _looks_like_email_document, _sniff_doc_language
+                from src.agent_tools.document_tools import _looks_like_email_document, _sniff_doc_language
                 language = _sniff_doc_language(req.content)
             else:
-                from src.tool_implementations import _looks_like_email_document
+                from src.agent_tools.document_tools import _looks_like_email_document
             if _looks_like_email_document(req.content, req.title):
                 language = "email"
 
@@ -171,11 +189,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         if session_id:
             db = SessionLocal()
             try:
-                sess = db.query(DbSession).filter(DbSession.id == session_id).first()
-                if not sess:
-                    raise HTTPException(404, "Session not found")
-                if user and sess.owner and sess.owner != user:
-                    raise HTTPException(403, "Cannot import into another user's session")
+                _get_session_or_404(db, session_id, user)
             finally:
                 db.close()
 
@@ -198,7 +212,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
 
         title = os.path.splitext(meta.get("original_name") or meta.get("name") or upload_id)[0]
         try:
-            body_text = strip_pdf_content_marker(_process_pdf(pdf_path))
+            body_text = strip_pdf_content_marker(_process_pdf(pdf_path, owner=user))
         except Exception:
             body_text = None
 
@@ -260,18 +274,29 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         db = SessionLocal()
         try:
             from sqlalchemy import or_
+            pdf_marker_cond = or_(
+                Document.current_content.like('%<!-- pdf_source upload_id="%'),
+                Document.current_content.like('%<!-- pdf_form_source upload_id="%'),
+            )
+            library_language_expr = case(
+                (pdf_marker_cond, "pdf"),
+                (Document.language.is_(None), "text"),
+                else_=Document.language,
+            )
             # Archived view shows ONLY archived docs; the default view excludes
             # them (NULL = legacy rows that predate the column = not archived).
             _arch_cond = (Document.archived == True) if archived else or_(
                 Document.archived == False, Document.archived.is_(None))
-            # Language facet counts (owner-filtered)
+            # Language facet counts (owner-filtered). PDF documents are stored
+            # as markdown wrappers, so group by the library display language
+            # instead of the raw stored language.
             lang_q = (
-                db.query(Document.language, func.count(Document.id))
+                db.query(library_language_expr, func.count(Document.id))
                 .outerjoin(DbSession, Document.session_id == DbSession.id)
                 .filter(Document.is_active == True).filter(_arch_cond)
             )
             lang_q = _owner_session_filter(lang_q, user)
-            lang_rows = lang_q.group_by(Document.language).all()
+            lang_rows = lang_q.group_by(library_language_expr).all()
             languages = _aggregate_language_facets(lang_rows)
 
             # Session count (owner-filtered)
@@ -303,12 +328,17 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                         Document.title.ilike(term) | Document.current_content.ilike(term)
                     )
 
-            # Language filter
+            # Language filter. "pdf" is a display language derived from the
+            # source marker; "markdown" excludes those wrappers.
             if language:
                 if language == "text":
                     q = q.filter((Document.language == None) | (Document.language == "text"))
+                elif language == "pdf":
+                    q = q.filter(pdf_marker_cond)
                 else:
                     q = q.filter(Document.language == language)
+                    if language == "markdown":
+                        q = q.filter(~pdf_marker_cond)
 
             # Total before pagination
             total = q.count()
@@ -332,7 +362,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                     "session_id": doc.session_id,
                     "session_name": session_name,
                     "title": doc.title,
-                    "language": doc.language or "text",
+                    "language": _library_language_for_document(doc),
                     "preview": (doc.current_content or "")[:500],
                     "version_count": doc.version_count,
                     "created_at": (doc.created_at.isoformat() + "Z") if doc.created_at else None,
@@ -359,18 +389,17 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         try:
             if not user:
                 raise HTTPException(403, "Authentication required")
-            session = db.query(DbSession).filter(DbSession.id == session_id).first()
             # v2 review HIGH-9: raise 403 explicitly when the caller
             # can't see this session, instead of returning [] which the
             # UI treats identically to "no docs" and silently masks
             # auth failures.
-            if not session:
-                raise HTTPException(404, "Session not found")
-            if user and session.owner and session.owner != user:
-                raise HTTPException(403, "Access denied")
-            docs = db.query(Document).filter(
+            _get_session_or_404(db, session_id, user)
+            q = db.query(Document).filter(
                 Document.session_id == session_id
-            ).order_by(Document.created_at.desc()).all()
+            )
+            if user:
+                q = q.filter(or_(Document.owner == user, Document.owner.is_(None)))
+            docs = q.order_by(Document.created_at.desc()).all()
             return [_doc_to_dict(d) for d in docs]
         finally:
             db.close()
@@ -437,7 +466,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                 raise HTTPException(404, "Source PDF could not be located")
 
             try:
-                body_text = strip_pdf_content_marker(_process_pdf(pdf_path))
+                body_text = strip_pdf_content_marker(_process_pdf(pdf_path, owner=user))
             except Exception as e:
                 logger.error(f"extract_pdf_text failed for {pdf_path}: {e}")
                 raise HTTPException(500, f"Extraction failed: {e}")
@@ -606,13 +635,15 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                 doc.language = req.language
             if req.session_id is not None:
                 # Empty string = unlink from session
+                if req.session_id:
+                    _get_session_or_404(db, req.session_id, user)
                 doc.session_id = req.session_id if req.session_id else None
                 if not req.session_id:
                     # Tab closed / doc detached from its session — drop the
                     # in-memory active-doc pointer so the last-resort injection
                     # path doesn't re-surface this doc in a later chat (#1160).
                     try:
-                        from src.tool_implementations import clear_active_document
+                        from src.agent_tools.document_tools import clear_active_document
                         clear_active_document(doc_id)
                     except Exception:
                         pass
@@ -641,7 +672,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             # Closed/deleted — drop the in-memory active-doc pointer so it isn't
             # re-injected into a later, unrelated chat (#1160).
             try:
-                from src.tool_implementations import clear_active_document
+                from src.agent_tools.document_tools import clear_active_document
                 clear_active_document(doc_id)
             except Exception:
                 pass
@@ -663,8 +694,9 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         try:
             # Verify ownership before listing versions
             doc = db.query(Document).filter(Document.id == doc_id).first()
-            if doc:
-                _verify_doc_owner(db, doc, user)
+            if not doc:
+                raise HTTPException(404, "Document not found")
+            _verify_doc_owner(db, doc, user)
             versions = db.query(DocumentVersion).filter(
                 DocumentVersion.document_id == doc_id
             ).order_by(DocumentVersion.version_number.desc()).all()
@@ -687,8 +719,9 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         try:
             # Verify ownership
             doc = db.query(Document).filter(Document.id == doc_id).first()
-            if doc:
-                _verify_doc_owner(db, doc, user)
+            if not doc:
+                raise HTTPException(404, "Document not found")
+            _verify_doc_owner(db, doc, user)
             ver = db.query(DocumentVersion).filter(
                 DocumentVersion.document_id == doc_id,
                 DocumentVersion.version_number == num,
@@ -853,10 +886,10 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         from src.llm_core import llm_call_async
 
         user = get_current_user(request)
-        url, model, headers = resolve_task_endpoint()
+        url, model, headers = resolve_task_endpoint(owner=user or None)
         if not url or not model:
             # Fall back to default endpoint
-            url, model, headers = resolve_endpoint("default")
+            url, model, headers = resolve_endpoint("default", owner=user or None)
         if not url or not model:
             raise HTTPException(500, "No endpoint configured for AI tidy")
 
@@ -1156,7 +1189,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         settings = _load_vl_settings()
         vl_model = settings.get("vision_model", "")
         try:
-            url, model_id, headers = _resolve_vl_model(vl_model)
+            url, model_id, headers = _resolve_vl_model(vl_model, owner=user)
         except Exception as e:
             raise HTTPException(503, f"No vision model available: {e}")
 
@@ -1510,10 +1543,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
         # don't import from a routes file (cycle-prone). Same env override
         # as email_routes (ODYSSEUS_MAIL_ATTACHMENTS_DIR).
         from pathlib import Path as _Path
-        import os as _os
-        _DATA_DIR = _Path(__file__).resolve().parent.parent / "data"
-        _BASE = _os.environ.get("ODYSSEUS_MAIL_ATTACHMENTS_DIR", str(_DATA_DIR / "mail-attachments"))
-        _COMPOSE_DIR = _Path(_BASE) / "_compose"
+        _COMPOSE_DIR = _Path(MAIL_ATTACHMENTS_DIR) / "_compose"
         _COMPOSE_DIR.mkdir(parents=True, exist_ok=True)
 
         user = get_current_user(request)
@@ -1629,9 +1659,11 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             #    context (To/Subject/In-Reply-To/References).
             try:
                 from routes.email_routes import _imap, _decode_header
+                from routes.email_helpers import _q
             except Exception:
                 _imap = None
                 _decode_header = lambda x: x or ""
+                _q = lambda x: x or ""
 
             to_addr = ""
             from_name = ""
@@ -1641,7 +1673,7 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             if _imap:
                 try:
                     with _imap(doc.source_email_account_id or None) as conn:
-                        conn.select(doc.source_email_folder, readonly=True)
+                        conn.select(_q(doc.source_email_folder), readonly=True)
                         status, data = conn.fetch(doc.source_email_uid.encode(), "(RFC822.HEADER)")
                     if status == "OK" and data and data[0]:
                         raw_hdr = data[0][1]

routes/email_helpers.py

@@ -71,6 +71,38 @@ def _send_smtp_message(cfg: dict, from_addr: str, recipients: list[str], message
         smtp.sendmail(from_addr, recipients, message)
 
 
+def _friendly_email_auth_error(protocol: str, host: str, error: object) -> str:
+    """Return a clearer setup error for known provider auth policies."""
+    raw = str(error or "")
+    lower = raw.lower()
+    host_lower = (host or "").lower()
+    microsoft_host = any(
+        marker in host_lower
+        for marker in (
+            "outlook.office365.com",
+            "smtp.office365.com",
+            "office365.com",
+            "outlook.com",
+            "hotmail.com",
+            "live.com",
+        )
+    )
+    microsoft_basic_auth_failure = (
+        "5.7.139" in lower
+        or "basic authentication is disabled" in lower
+        or ("authenticate failed" in lower and microsoft_host)
+        or ("authentication unsuccessful" in lower and microsoft_host)
+    )
+    if microsoft_basic_auth_failure:
+        return (
+            "Microsoft no longer accepts normal mailbox passwords for "
+            "Outlook/Office 365 IMAP/SMTP in most accounts. Odysseus "
+            "does not support Microsoft OAuth/Graph mail yet, so Outlook "
+            "accounts cannot be added with this password form."
+        )
+    return raw[:200]
+
+
 def _strip_think(text: str) -> str:
     """Email-flavored think strip — thin wrapper over the central helper.
 
@@ -254,16 +286,17 @@ def _cleanup_compose_uploads(tokens) -> None:
             pass
 
 
-DATA_DIR = Path(__file__).resolve().parent.parent / "data"
-SETTINGS_FILE = DATA_DIR / "settings.json"
+from src.constants import DATA_DIR as _DATA_DIR, MAIL_ATTACHMENTS_DIR, SETTINGS_FILE as _SETTINGS_FILE, SCHEDULED_EMAILS_DB
+DATA_DIR = Path(_DATA_DIR)
+SETTINGS_FILE = Path(_SETTINGS_FILE)
 # Override at deploy time via ODYSSEUS_MAIL_ATTACHMENTS_DIR. Defaults to a
 # subdir of the install's data/ tree so the app works out-of-the-box without
 # a hardcoded /home/<user>/ path.
-ATTACHMENTS_DIR = Path(os.environ.get("ODYSSEUS_MAIL_ATTACHMENTS_DIR", str(DATA_DIR / "mail-attachments")))
+ATTACHMENTS_DIR = Path(MAIL_ATTACHMENTS_DIR)
 ATTACHMENTS_DIR.mkdir(parents=True, exist_ok=True)
 COMPOSE_UPLOADS_DIR = ATTACHMENTS_DIR / "_compose"
 COMPOSE_UPLOADS_DIR.mkdir(parents=True, exist_ok=True)
-SCHEDULED_DB = DATA_DIR / "scheduled_emails.db"
+SCHEDULED_DB = Path(SCHEDULED_EMAILS_DB)
 
 
 OWNER_SCOPED_EMAIL_CACHE_TABLES = {
@@ -271,6 +304,7 @@ OWNER_SCOPED_EMAIL_CACHE_TABLES = {
     "email_ai_replies",
     "email_calendar_extractions",
     "email_urgency_alerts",
+    "sender_signatures",
 }
 
 
@@ -308,6 +342,55 @@ def _ensure_owner_scoped_email_cache_table(conn, table: str, create_sql: str, co
         _lg.getLogger(__name__).warning(f"{table} owner-migration skipped: {_mig_e}")
 
 
+def _ensure_sender_signatures_table(conn):
+    """Create/migrate learned sender signatures to an owner-scoped cache."""
+    create_sql = """
+        CREATE TABLE IF NOT EXISTS sender_signatures (
+            from_address TEXT,
+            owner TEXT DEFAULT '',
+            signature_text TEXT,
+            sample_count INTEGER,
+            last_built_at TEXT NOT NULL,
+            model_used TEXT,
+            source TEXT,
+            PRIMARY KEY (from_address, owner)
+        )
+    """
+    conn.execute(create_sql)
+    try:
+        info = conn.execute("PRAGMA table_info(sender_signatures)").fetchall()
+        cols = [r[1] for r in info]
+        pk_cols = [r[1] for r in sorted((r for r in info if r[5]), key=lambda r: r[5])]
+        if "owner" in cols and pk_cols == ["from_address", "owner"]:
+            return
+
+        conn.execute("ALTER TABLE sender_signatures RENAME TO sender_signatures__old")
+        conn.execute(create_sql)
+        old_cols = [r[1] for r in conn.execute("PRAGMA table_info(sender_signatures__old)").fetchall()]
+        copy_cols = [
+            c for c in (
+                "from_address",
+                "signature_text",
+                "sample_count",
+                "last_built_at",
+                "model_used",
+                "source",
+            )
+            if c in old_cols
+        ]
+        source_owner = "COALESCE(owner, '')" if "owner" in old_cols else "''"
+        conn.execute(
+            f"INSERT OR IGNORE INTO sender_signatures "
+            f"({', '.join([*copy_cols, 'owner'])}) "
+            f"SELECT {', '.join([*copy_cols, source_owner])} "
+            f"FROM sender_signatures__old"
+        )
+        conn.execute("DROP TABLE sender_signatures__old")
+    except Exception as _mig_e:
+        import logging as _lg
+        _lg.getLogger(__name__).warning(f"sender_signatures owner-migration skipped: {_mig_e}")
+
+
 def attachment_extract_dir(folder: str, uid: str) -> Path:
     """Containment-safe extraction directory for an attachment.
 
@@ -526,20 +609,10 @@ def _init_scheduled_db():
             conn.execute("ALTER TABLE email_boundaries ADD COLUMN turns_json TEXT")
     except Exception:
         pass
-    # Per-sender signature cache. Populated by `learn_sender_signatures`
-    # action: the LLM extracts the common trailing block across N emails
-    # from each sender; the renderer folds it consistently for every
-    # future email from that address.
-    conn.execute("""
-        CREATE TABLE IF NOT EXISTS sender_signatures (
-            from_address TEXT PRIMARY KEY,
-            signature_text TEXT,
-            sample_count INTEGER,
-            last_built_at TEXT NOT NULL,
-            model_used TEXT,
-            source TEXT
-        )
-    """)
+    # Per-sender signature cache. Populated by `learn_sender_signatures`.
+    # Message sender addresses are global, so signatures must be scoped to the
+    # mailbox owner before `/read` returns them to the renderer.
+    _ensure_sender_signatures_table(conn)
     conn.commit()
     conn.close()
 
@@ -705,7 +778,16 @@ def _open_imap_connection(host: str, port: int, *, starttls: bool, timeout: int
     port = int(port or 993)
     if starttls:
         conn = imaplib.IMAP4(host, port, timeout=timeout)
-        conn.starttls()
+        try:
+            conn.starttls()
+        except Exception:
+            # Don't leak the open plain socket if the STARTTLS upgrade is
+            # rejected; close it before propagating. (#3174)
+            try:
+                conn.shutdown()
+            except Exception:
+                pass
+            raise
     elif port == 993:
         conn = imaplib.IMAP4_SSL(host, port, timeout=timeout)
     else:
@@ -714,12 +796,20 @@ def _open_imap_connection(host: str, port: int, *, starttls: bool, timeout: int
         conn.sock.settimeout(timeout)
     except Exception:
         pass
+    # Raise the IMAP line-length limit from the default 1 MB to 50 MB so that
+    # large mailboxes (tens of thousands of messages) don't crash with
+    # "got more than 1000000 bytes" on UID SEARCH ALL.  (#2883)
+    imaplib._MAXLINE = 50_000_000
     return conn
 
-def _imap_connect(account_id: str | None = None, owner: str = ""):
+def _imap_connect(account_id: str | None = None, owner: str = "",
+                  timeout: int = _IMAP_TIMEOUT_SECONDS):
     # SECURITY: passing `owner` scopes the fallback config lookup so a brand
     # new user doesn't get connected against another user's default mailbox
     # when they have no account configured.
+    #
+    # `timeout` is overridable so short-lived callers (e.g. the service-health
+    # probe) can impose a tighter budget than the default IMAP timeout.
     cfg = _get_email_config(account_id, owner=owner)
     # Connection mode:
     #   STARTTLS on → plain + upgrade
@@ -732,9 +822,20 @@ def _imap_connect(account_id: str | None = None, owner: str = ""):
         cfg["imap_host"],
         cfg["imap_port"],
         starttls=bool(cfg.get("imap_starttls")),
-        timeout=_IMAP_TIMEOUT_SECONDS,
+        timeout=timeout,
     )
-    conn.login(cfg["imap_user"], cfg["imap_password"])
+    try:
+        conn.login(cfg["imap_user"], cfg["imap_password"])
+    except Exception:
+        # A failed AUTHENTICATE (e.g. an Office 365 app password on an
+        # MFA-enabled tenant, #3174) otherwise orphans the already-connected
+        # socket; close it before propagating so a misconfigured account
+        # can't leak one descriptor per retry / background poller pass.
+        try:
+            conn.shutdown()
+        except Exception:
+            pass
+        raise
     return conn
 
 
@@ -798,20 +899,28 @@ def _imap(account_id: str | None = None, owner: str = ""):
 def _decode_header(raw):
     if not raw:
         return ""
-    parts = email.header.decode_header(raw)
-    decoded = []
-    for data, charset in parts:
-        if isinstance(data, bytes):
-            try:
-                decoded.append(data.decode(charset or "utf-8", errors="replace"))
-            except (LookupError, ValueError):
-                # Unknown/invalid MIME charset (e.g. a malformed or spam header
-                # like =?x-unknown-charset?B?...?=). errors="replace" only covers
-                # byte-decode errors, not codec lookup, so fall back to utf-8.
-                decoded.append(data.decode("utf-8", errors="replace"))
-        else:
-            decoded.append(data)
-    return " ".join(decoded)
+    try:
+        # make_header concatenates per RFC 2047: no spurious space between an
+        # encoded-word and adjacent plain text (plain runs keep their own
+        # whitespace), and the whitespace between two adjacent encoded-words is
+        # dropped. The old " ".join produced "Re:  Jose"-style double spaces on
+        # every non-ASCII subject or sender.
+        return str(email.header.make_header(email.header.decode_header(raw)))
+    except Exception:
+        # Malformed header or unknown/invalid MIME charset (e.g. a spam header
+        # like =?x-unknown-charset?B?...?=) makes make_header raise LookupError;
+        # fall back to a lossy per-part decode. errors="replace" only covers
+        # byte-decode errors, not codec lookup, hence the explicit utf-8 retry.
+        decoded = []
+        for data, charset in email.header.decode_header(raw):
+            if isinstance(data, bytes):
+                try:
+                    decoded.append(data.decode(charset or "utf-8", errors="replace"))
+                except (LookupError, ValueError):
+                    decoded.append(data.decode("utf-8", errors="replace"))
+            else:
+                decoded.append(data)
+        return "".join(decoded)
 
 
 def _detect_sent_folder(conn):
@@ -1136,13 +1245,9 @@ def _fetch_sender_thread_context(sender_addr: str,
     if exclude_uid:
         seen_uids.add((exclude_folder or "INBOX", str(exclude_uid)))
 
+    conn = None
     try:
         conn = _imap_connect(account_id, owner=owner)
-    except Exception as e:
-        logger.warning(f"sender-thread-context: imap connect failed: {e}")
-        return ""
-
-    try:
         for folder in ["INBOX", "Sent", "Archive", "Drafts"]:
             if len(blocks) >= limit:
                 break
@@ -1209,11 +1314,14 @@ def _fetch_sender_thread_context(sender_addr: str,
                 if atts_text:
                     lines.append(atts_text)
                 blocks.append("\n".join(lines))
+    except Exception as e:
+        logger.warning(f"sender-thread-context: imap failed: {e}")
     finally:
-        try: conn.close()
-        except Exception: pass
-        try: conn.logout()
-        except Exception: pass
+        if conn:
+            try: conn.close()
+            except Exception: pass
+            try: conn.logout()
+            except Exception: pass
 
     if not blocks:
         return ""
@@ -1316,6 +1424,7 @@ def _pre_retrieve_context(
         if not terms_list:
             return context_snippets, terms_list
 
+        ctx_conn = None
         try:
             ctx_conn = _imap_connect(account_id, owner=owner)
             for folder in ["INBOX", "Sent", "Archive", "Drafts"]:
@@ -1352,12 +1461,12 @@ def _pre_retrieve_context(
                     except Exception as _e:
                         logger.warning(f"  search {folder} {term!r} failed: {_e}")
                         continue
-            try:
-                ctx_conn.logout()
-            except Exception:
-                pass
         except Exception as _e:
             logger.warning(f"IMAP context search failed: {_e}")
+        finally:
+            if ctx_conn:
+                try: ctx_conn.logout()
+                except Exception: pass
 
         try:
             from routes.contacts_routes import _fetch_contacts

routes/email_pollers.py

@@ -210,7 +210,7 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
         if auto_cal:
             for sent_name in ("Sent", "INBOX/Sent", "Sent Items", "[Gmail]/Sent Mail"):
                 try:
-                    st, _ = conn.select(sent_name, readonly=True)
+                    st, _ = conn.select(_q(sent_name), readonly=True)
                     if st == "OK":
                         folders_to_scan.append(sent_name)
                         break
@@ -1046,7 +1046,7 @@ def _scheduled_poll_once() -> dict:
                 try:
                     with _imap(row_account_id, owner=row_owner) as imap:
                         sent_folder = _detect_sent_folder(imap)
-                        imap.append(sent_folder, "\\Seen", None, outer.as_bytes())
+                        imap.append(_q(sent_folder), "\\Seen", None, outer.as_bytes())
                 except Exception as e:
                     logger.warning(f"Failed to append scheduled {sid} to Sent: {e}")
 

routes/email_routes.py

@@ -32,9 +32,10 @@ from email.mime.multipart import MIMEMultipart
 
 from fastapi import APIRouter, Query, UploadFile, File, BackgroundTasks, HTTPException, Depends, Request
 from fastapi.responses import FileResponse
+from src.constants import DATA_DIR
 
 from src.llm_core import llm_call_async
-from src.upload_limits import read_upload_limited
+from src.upload_limits import read_upload_limited, EMAIL_COMPOSE_UPLOAD_MAX_BYTES
 
 from routes.email_helpers import (
     _strip_think, _extract_reply, _apply_email_style_mechanics, require_owner, require_user, _assert_owns_account,
@@ -47,6 +48,7 @@ from routes.email_helpers import (
     _extract_attachment_to_disk, _extract_html, _extract_text,
     _fetch_sender_thread_context, _pre_retrieve_context,
     _EMAIL_REPLY_SYS_PROMPT_BASE, _POOL_HOOKS,
+    _friendly_email_auth_error,
     SendEmailRequest, ExtractStyleRequest,
     ATTACHMENTS_DIR, COMPOSE_UPLOADS_DIR, SCHEDULED_DB,
     attachment_extract_dir, _email_cache_owner_clause,
@@ -56,7 +58,6 @@ from routes.email_pollers import _start_poller
 logger = logging.getLogger(__name__)
 
 ODYSSEUS_MAIL_ORIGIN = "odysseus-ui"
-EMAIL_COMPOSE_UPLOAD_MAX_BYTES = 25 * 1024 * 1024
 
 
 def _email_tag_owner_aliases(account_id: str | None, owner: str = "") -> list[str]:
@@ -248,6 +249,41 @@ def _uid_from_fetch_meta(meta_b: bytes) -> str:
     return m.group(1).decode() if m else ""
 
 
+_FETCH_SEQ_RE = re.compile(rb"^(\d+)\s+\(")
+
+
+def _group_uid_fetch_records(msg_data) -> list:
+    """Group an imaplib UID FETCH response into per-message (meta, payload).
+
+    imaplib yields an interleaved list: ``(meta, literal)`` tuples for
+    attributes that carry a literal (``RFC822.HEADER {n}`` etc.) plus bare
+    ``bytes`` elements for everything the server sends outside a literal.
+    Where each attribute lands is server-specific: Dovecot sends FLAGS
+    *before* the header literal (so it ends up inside the tuple meta), while
+    Gmail sends FLAGS *after* it, arriving as a bare ``b' FLAGS (\\Seen))'``
+    element. Dropping bare elements therefore silently loses FLAGS on Gmail
+    and every message renders as unread/unflagged.
+
+    A tuple whose meta starts with a sequence number opens a new record;
+    every other part — continuation tuple or bare bytes — is folded into the
+    current record's meta so attribute regexes see the full meta text.
+    Plain ``b')'`` terminators get folded in too, which is harmless.
+    """
+    grouped: list = []  # list of (meta_bytes, payload_bytes_or_None)
+    for part in (msg_data or []):
+        if isinstance(part, tuple):
+            meta_b = part[0] if isinstance(part[0], (bytes, bytearray)) else str(part[0]).encode()
+            if _FETCH_SEQ_RE.match(meta_b):
+                grouped.append((meta_b, part[1]))
+            elif grouped:
+                cur_meta, cur_payload = grouped[-1]
+                grouped[-1] = (cur_meta + b" " + meta_b, cur_payload or part[1])
+        elif isinstance(part, (bytes, bytearray)) and grouped:
+            cur_meta, cur_payload = grouped[-1]
+            grouped[-1] = (cur_meta + b" " + bytes(part), cur_payload)
+    return grouped
+
+
 def _smtp_ready(cfg: dict) -> bool:
     return bool(cfg.get("smtp_host") and cfg.get("smtp_user") and cfg.get("smtp_password"))
 
@@ -798,20 +834,11 @@ def setup_email_routes():
                 except Exception as e:
                     logger.warning(f"Batch fetch failed, falling back to per-UID: {e}")
                     status, msg_data = "NO", []
-                # imaplib batch responses interleave (meta, payload) tuples and
-                # `b')'` terminators. Group by message: each tuple where the
-                # meta begins with a seq number starts a new message record.
-                seq_re = re.compile(rb'^(\d+)\s+\(')
-                grouped = []  # list of (meta_str, payload_bytes)
-                for part in (msg_data or []):
-                    if isinstance(part, tuple):
-                        meta_b = part[0] if isinstance(part[0], (bytes, bytearray)) else str(part[0]).encode()
-                        if seq_re.match(meta_b):
-                            grouped.append((meta_b, part[1]))
-                        elif grouped:
-                            # continuation of previous message — concatenate meta info if any
-                            cur_meta, cur_payload = grouped[-1]
-                            grouped[-1] = (cur_meta + b" " + meta_b, cur_payload or part[1])
+                # Group the batched response into per-message (meta, payload)
+                # records. Bare bytes parts must be kept: Gmail returns FLAGS
+                # after the header literal as a bare element, and dropping it
+                # rendered every Gmail message as unread/unflagged.
+                grouped = _group_uid_fetch_records(msg_data)
 
                 if status != "OK" and not grouped:
                     conn.logout()
@@ -1097,14 +1124,15 @@ def setup_email_routes():
                             continue
                         raw_header = None
                         flags = ""
-                        for part in msg_data:
-                            if isinstance(part, tuple):
-                                meta = part[0].decode() if isinstance(part[0], bytes) else str(part[0])
-                                if b"RFC822.HEADER" in part[0] if isinstance(part[0], bytes) else "RFC822.HEADER" in meta:
-                                    raw_header = part[1]
-                                flag_match = re.search(r'FLAGS \(([^)]*)\)', meta)
-                                if flag_match:
-                                    flags = flag_match.group(1)
+                        # Same Gmail caveat as the list route: FLAGS may
+                        # arrive after the header literal, so group bare
+                        # parts back into the message meta before scanning.
+                        for meta_b, payload in _group_uid_fetch_records(msg_data):
+                            if payload and b"RFC822.HEADER" in meta_b:
+                                raw_header = payload
+                            flag_match = re.search(rb'FLAGS \(([^)]*)\)', meta_b)
+                            if flag_match:
+                                flags = flag_match.group(1).decode(errors="replace")
                         if not raw_header:
                             continue
                         msg = email_mod.message_from_bytes(raw_header)
@@ -1246,8 +1274,9 @@ def setup_email_routes():
                 try:
                     if sender_addr:
                         _rs = _c.execute(
-                            "SELECT signature_text FROM sender_signatures WHERE from_address = ?",
-                            (sender_addr.lower().strip(),),
+                            f"SELECT signature_text FROM sender_signatures "
+                            f"WHERE from_address = ? AND {owner_clause}",
+                            (sender_addr.lower().strip(), *owner_params),
                         ).fetchone()
                         if _rs and _rs[0]:
                             cached_sender_sig = _rs[0]
@@ -2904,7 +2933,7 @@ def setup_email_routes():
         from pathlib import Path as _P
         import json as _json
         _slug = "".join(c if (c.isalnum() or c in "-_.@") else "_" for c in (owner or "default"))
-        path = _P(f"data/email_urgency_state_{_slug}.json")
+        path = _P(DATA_DIR) / f"email_urgency_state_{_slug}.json"
         if not path.exists():
             return {"total_unread": 0, "total_urgent": 0, "max_score": 0, "per_uid": {}}
         try:
@@ -3162,7 +3191,7 @@ def setup_email_routes():
                     try: conn.logout()
                     except Exception: pass
             except Exception as e:
-                imap_result = {"ok": False, "error": str(e)[:200]}
+                imap_result = {"ok": False, "error": _friendly_email_auth_error("IMAP", imap_host, e)}
 
         smtp_host = (body.get("smtp_host") or "").strip()
         if smtp_host:
@@ -3184,7 +3213,7 @@ def setup_email_routes():
                     try: smtp.quit()
                     except Exception: pass
             except Exception as e:
-                smtp_result = {"ok": False, "error": str(e)[:200]}
+                smtp_result = {"ok": False, "error": _friendly_email_auth_error("SMTP", smtp_host, e)}
 
         return {
             "ok": imap_result["ok"] and (smtp_result is None or smtp_result["ok"]),

routes/embedding_routes.py

@@ -7,12 +7,12 @@ import logging
 import asyncio
 from pathlib import Path
 from fastapi import APIRouter, HTTPException, Form, Depends
-from core.constants import BASE_DIR
+from core.constants import EMBEDDING_ENDPOINT_FILE, FASTEMBED_CACHE_DIR
 from core.middleware import require_admin
 
 logger = logging.getLogger(__name__)
 
-_ENDPOINT_FILE = os.path.join(BASE_DIR, "data", "embedding_endpoint.json")
+_ENDPOINT_FILE = EMBEDDING_ENDPOINT_FILE
 
 # Track in-progress downloads
 _downloading: dict = {}
@@ -35,13 +35,7 @@ def _cache_dir() -> str:
     default lived in /tmp, which many systems wipe on reboot — forcing a
     full re-download of the embedding model after every restart.
     """
-    env = os.environ.get("FASTEMBED_CACHE_PATH")
-    if env:
-        return env
-    return os.path.join(
-        os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
-        "data", "fastembed_cache",
-    )
+    return FASTEMBED_CACHE_DIR
 
 
 def _model_cache_name(hf_source: str) -> str:
@@ -49,19 +43,35 @@ def _model_cache_name(hf_source: str) -> str:
     return "models--" + hf_source.replace("/", "--")
 
 
+def _model_cache_path(hf_source: str) -> Path:
+    """Return a confined cache path for a fastembed HF source."""
+    root = Path(_cache_dir()).expanduser().resolve()
+    raw_path = root / _model_cache_name(hf_source)
+    if raw_path.is_symlink():
+        raise ValueError("Model cache path must not be a symlink")
+    path = raw_path.resolve(strict=False)
+    try:
+        path.relative_to(root)
+    except ValueError:
+        raise ValueError("Model cache path escapes cache root")
+    return path
+
+
 def _is_downloaded(hf_source: str) -> bool:
     """Check if a model is already cached."""
-    cache = _cache_dir()
-    model_dir = os.path.join(cache, _model_cache_name(hf_source))
-    if not os.path.isdir(model_dir):
+    try:
+        model_dir = _model_cache_path(hf_source)
+    except ValueError:
+        return False
+    if not model_dir.is_dir():
         return False
     # Check for actual model files (not just empty dir)
-    snapshots = os.path.join(model_dir, "snapshots")
-    if os.path.isdir(snapshots):
-        return any(os.listdir(snapshots))
+    snapshots = model_dir / "snapshots"
+    if snapshots.is_dir():
+        return any(snapshots.iterdir())
     # Also check for blobs (older cache format)
-    blobs = os.path.join(model_dir, "blobs")
-    return os.path.isdir(blobs) and any(os.listdir(blobs))
+    blobs = model_dir / "blobs"
+    return blobs.is_dir() and any(blobs.iterdir())
 
 
 def _active_model() -> str:
@@ -119,8 +129,10 @@ def setup_embedding_routes():
 
             cached_size = None
             if downloaded and hf_src:
-                model_path = os.path.join(_cache_dir(), _model_cache_name(hf_src))
-                cached_size = _dir_size_mb(model_path)
+                try:
+                    cached_size = _dir_size_mb(str(_model_cache_path(hf_src)))
+                except ValueError:
+                    cached_size = None
 
             result.append({
                 "model": m["model"],
@@ -217,8 +229,11 @@ def setup_embedding_routes():
         if not hf_src:
             raise HTTPException(400, "No cache source for this model")
 
-        model_path = os.path.join(_cache_dir(), _model_cache_name(hf_src))
-        if not os.path.isdir(model_path):
+        try:
+            model_path = _model_cache_path(hf_src)
+        except ValueError as e:
+            raise HTTPException(400, str(e))
+        if not model_path.is_dir():
             return {"deleted": False, "message": "Model not cached"}
 
         shutil.rmtree(model_path)
@@ -237,7 +252,7 @@ def setup_embedding_routes():
         }
 
     @router.post("/endpoint")
-    def set_endpoint(url: str = Form(...), model: str = Form("")):
+    def set_endpoint(url: str = Form(...), model: str = Form(""), api_key: str = Form("")):
         """Save a custom embedding endpoint URL."""
         url = url.strip()
         if not url:
@@ -261,6 +276,7 @@ def setup_embedding_routes():
             resp = httpx.post(
                 url,
                 json={"input": ["test"], "model": model or "test"},
+                headers={"Authorization": f"Bearer {api_key}"} if api_key else {},
                 timeout=10,
             )
             resp.raise_for_status()
@@ -271,10 +287,16 @@ def setup_embedding_routes():
         data = {"url": url}
         if model:
             data["model"] = model
+        if api_key:
+            from src.secret_storage import encrypt
+            data["api_key"] = encrypt(api_key)
+
         _save_custom_endpoint(data)
         os.environ["EMBEDDING_URL"] = url
         if model:
             os.environ["EMBEDDING_MODEL"] = model
+        if api_key:
+            os.environ["EMBEDDING_API_KEY"] = api_key
 
         # Reset the RAG singleton so it picks up the new endpoint
         import src.rag_singleton as _rs
@@ -288,6 +310,16 @@ def setup_embedding_routes():
             reset_http_embed_state()
         except Exception:
             pass
+        try:
+            from src.embedding_lanes import reset_embedding_lane_state
+            reset_embedding_lane_state()
+        except Exception:
+            pass
+        try:
+            from src.tool_index import reset_tool_index
+            reset_tool_index()
+        except Exception:
+            pass
 
         # Reset ChromaDB client (collections will be recreated with new embeddings)
         try:
@@ -308,6 +340,7 @@ def setup_embedding_routes():
         # Remove from environment
         os.environ.pop("EMBEDDING_URL", None)
         os.environ.pop("EMBEDDING_MODEL", None)
+        os.environ.pop("EMBEDDING_API_KEY", None)
 
         # Reset the RAG singleton so it falls back to fastembed
         import src.rag_singleton as _rs
@@ -318,6 +351,16 @@ def setup_embedding_routes():
             reset_http_embed_state()
         except Exception:
             pass
+        try:
+            from src.embedding_lanes import reset_embedding_lane_state
+            reset_embedding_lane_state()
+        except Exception:
+            pass
+        try:
+            from src.tool_index import reset_tool_index
+            reset_tool_index()
+        except Exception:
+            pass
 
         # Reset ChromaDB client
         try:

routes/emoji_routes.py

@@ -16,22 +16,54 @@ from pathlib import Path
 
 import httpx
 from fastapi import APIRouter
-from fastapi.responses import FileResponse, Response
+from fastapi.responses import Response
+
+from src.constants import EMOJI_CACHE_DIR
 
 logger = logging.getLogger(__name__)
 
-_CACHE_DIR = Path(__file__).resolve().parent.parent / "data" / "emoji_cache"
+_CACHE_DIR = Path(EMOJI_CACHE_DIR)
 # OpenMoji "black" set = monochrome line-art SVGs. Filenames are the codepoints
 # in UPPERCASE (FE0F dropped, same as we compute), '-' joined.
 _OPENMOJI_BASE = "https://cdn.jsdelivr.net/npm/openmoji@15.0.0/black/svg"
 # codepoints like "1f600" or "1f468-200d-1f469-200d-1f467" (lowercase hex, '-' joined)
 _CODE_RE = re.compile(r"^[0-9a-f]{2,6}(?:-[0-9a-f]{2,6})*$")
-_SVG_HEADERS = {"Cache-Control": "public, max-age=31536000, immutable"}
+_MAX_SVG_BYTES = 256 * 1024
+_BLOCKED_SVG_RE = re.compile(
+    br"<\s*(?:script|foreignObject|iframe|object|embed|image)\b|"
+    br"\bon[a-z0-9_-]+\s*=",
+    re.IGNORECASE,
+)
+_EXTERNAL_REF_RE = re.compile(
+    br"\b(?:href|xlink:href)\s*=\s*['\"](?:https?:|//|data:|javascript:)",
+    re.IGNORECASE,
+)
+_SVG_SECURITY_HEADERS = {
+    "X-Content-Type-Options": "nosniff",
+    "Content-Security-Policy": "sandbox",
+    "Cross-Origin-Resource-Policy": "same-origin",
+}
+_SVG_HEADERS = {
+    "Cache-Control": "public, max-age=31536000, immutable",
+    **_SVG_SECURITY_HEADERS,
+}
 # Returned when a codepoint is unknown/unreachable: an empty (transparent) SVG,
 # so the CSS mask renders nothing instead of a solid box. Not cached, so a later
 # request can still pick up the real glyph once the CDN is reachable.
 _BLANK_SVG = b'<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1 1"></svg>'
-_BLANK_HEADERS = {"Cache-Control": "no-store"}
+_BLANK_HEADERS = {"Cache-Control": "no-store", **_SVG_SECURITY_HEADERS}
+
+
+def _is_safe_svg(content: bytes) -> bool:
+    if not isinstance(content, bytes) or not content:
+        return False
+    if len(content) > _MAX_SVG_BYTES:
+        return False
+    if b"<svg" not in content[:256].lower():
+        return False
+    if _BLOCKED_SVG_RE.search(content) or _EXTERNAL_REF_RE.search(content):
+        return False
+    return True
 
 
 def setup_emoji_routes() -> APIRouter:
@@ -49,14 +81,21 @@ def setup_emoji_routes() -> APIRouter:
         _CACHE_DIR.mkdir(parents=True, exist_ok=True)
         fp = _CACHE_DIR / f"{code}.svg"
         if fp.exists():
-            return FileResponse(fp, media_type="image/svg+xml", headers=_SVG_HEADERS)
+            try:
+                content = fp.read_bytes()
+                if _is_safe_svg(content):
+                    return Response(content, media_type="image/svg+xml", headers=_SVG_HEADERS)
+                fp.unlink(missing_ok=True)
+            except Exception as e:
+                logger.warning("emoji cache read %s failed: %s", code, e)
+            return _blank()
 
         # First time we've seen this emoji — fetch the OpenMoji black SVG + cache
         # it. OpenMoji filenames are the codepoints uppercased.
         try:
             async with httpx.AsyncClient(timeout=8.0) as client:
                 r = await client.get(f"{_OPENMOJI_BASE}/{code.upper()}.svg")
-            if r.status_code == 200 and b"<svg" in r.content[:256]:
+            if r.status_code == 200 and _is_safe_svg(r.content):
                 try:
                     fp.write_bytes(r.content)
                 except Exception:

routes/gallery_helpers.py

@@ -11,6 +11,7 @@ from typing import Dict, Any, Optional
 from pydantic import BaseModel
 
 from core.database import GalleryImage
+from src.auth_helpers import _auth_disabled
 
 logger = logging.getLogger(__name__)
 
@@ -120,19 +121,18 @@ def _image_to_dict(img: GalleryImage, session_name: str = None) -> Dict[str, Any
     }
 
 
-def _owner_filter(q, user):
+def _owner_filter(q, user, model_cls=GalleryImage):
     """Apply owner filtering to a gallery query.
 
-    When auth is disabled (single-user mode) get_current_user returns None
-    and there is no per-user scoping. The main library list and stats already
-    treat None as "show everything" (`if user is not None`), so this helper
-    must too — otherwise the tag/model filter sidebars come back empty and the
-    tag-cleanup endpoints (clear-user-tags, clear-ai-tags, dedupe-tags)
-    silently affect zero rows in the most common self-hosted deployment.
+    ``get_current_user`` returns None both in auth-disabled single-user mode
+    and when auth is enabled but no current user was resolved. Preserve the
+    single-user behavior, but fail closed for auth-enabled null-user states.
     """
-    if user is None:
+    if user is not None:
+        return q.filter(model_cls.owner == user)
+    if _auth_disabled():
         return q
-    return q.filter(GalleryImage.owner == user)
+    return q.filter(False)
 
 
 

routes/gallery_routes.py

@@ -12,8 +12,13 @@ from fastapi import APIRouter, HTTPException, Query, Request
 
 from core.database import SessionLocal, GalleryImage, GalleryAlbum, ModelEndpoint
 from core.database import Session as DbSession
-from src.auth_helpers import get_current_user, require_privilege
-from src.upload_limits import read_upload_limited
+from src.auth_helpers import get_current_user, owner_filter, require_privilege
+from src.upload_limits import (
+    read_upload_limited,
+    GALLERY_UPLOAD_MAX_BYTES,
+    GALLERY_TRANSFORM_UPLOAD_MAX_BYTES,
+)
+from src.constants import GENERATED_IMAGES_DIR
 
 from routes.gallery_helpers import (
     GalleryPatch, _extract_exif, _image_to_dict, _owner_filter, _human_size,
@@ -21,17 +26,88 @@ from routes.gallery_helpers import (
 
 logger = logging.getLogger(__name__)
 
-GALLERY_UPLOAD_MAX_BYTES = int(os.getenv("ODYSSEUS_GALLERY_UPLOAD_MAX_BYTES", str(100 * 1024 * 1024)))
-GALLERY_TRANSFORM_UPLOAD_MAX_BYTES = int(os.getenv("ODYSSEUS_GALLERY_TRANSFORM_UPLOAD_MAX_BYTES", str(25 * 1024 * 1024)))
+
+def _current_user_is_admin(request: Request, user: str | None) -> bool:
+    if not user:
+        return False
+    auth_mgr = getattr(request.app.state, "auth_manager", None)
+    is_admin = getattr(auth_mgr, "is_admin", None)
+    if not callable(is_admin):
+        return False
+    try:
+        return bool(is_admin(user))
+    except Exception:
+        return False
 
 
 def _sanitize_gallery_filename(filename: str) -> str:
     """Return a local filename safe to join under generated_images."""
-    safe_name = re.sub(r"[^A-Za-z0-9._-]", "_", Path(filename or "").name)[:128]
+    safe_name = re.sub(r"[^A-Za-z0-9._-]", "_", Path(str(filename or "")).name)[:128]
     if not safe_name or safe_name in {".", ".."}:
         safe_name = uuid.uuid4().hex[:12]
     return safe_name
 
+
+GALLERY_IMAGE_DIR = Path(GENERATED_IMAGES_DIR)
+
+
+def _gallery_image_path(filename: str) -> Path:
+    """Resolve a stored gallery filename without leaving generated_images."""
+    if not isinstance(filename, str):
+        raise HTTPException(400, "Unsafe gallery filename")
+    safe_name = _sanitize_gallery_filename(filename)
+    original = str(filename or "")
+    root = GALLERY_IMAGE_DIR.resolve()
+    path = (GALLERY_IMAGE_DIR / safe_name).resolve()
+    try:
+        if os.path.commonpath([str(root), str(path)]) != str(root):
+            raise ValueError
+    except Exception:
+        raise HTTPException(400, "Unsafe gallery filename")
+    if safe_name != original:
+        raise HTTPException(400, "Unsafe gallery filename")
+    return path
+
+
+def _normalize_image_endpoint_base(url: str) -> str:
+    base = (url or "").strip().rstrip("/")
+    if base.endswith("/v1"):
+        base = base[:-3].rstrip("/")
+    return base
+
+
+def _visible_image_endpoint_query(db, owner: str | None):
+    from src.auth_helpers import owner_filter
+    q = db.query(ModelEndpoint).filter(
+        ModelEndpoint.model_type == "image",
+        ModelEndpoint.is_enabled == True,  # noqa: E712
+    )
+    return owner_filter(q, ModelEndpoint, owner)
+
+
+def _first_visible_image_endpoint(db, owner: str | None):
+    endpoints = _visible_image_endpoint_query(db, owner).all()
+    if owner:
+        for ep in endpoints:
+            if getattr(ep, "owner", None) == owner:
+                return ep
+    return endpoints[0] if endpoints else None
+
+
+def _visible_image_endpoint_for_base(db, base: str, owner: str | None):
+    target = _normalize_image_endpoint_base(base)
+    if not target:
+        return None
+    fallback = None
+    for ep in _visible_image_endpoint_query(db, owner).all():
+        if _normalize_image_endpoint_base(getattr(ep, "base_url", "")) == target:
+            if owner and getattr(ep, "owner", None) == owner:
+                return ep
+            if fallback is None:
+                fallback = ep
+    return fallback
+
+
 def setup_gallery_routes() -> APIRouter:
     router = APIRouter(tags=["gallery"])
 
@@ -55,6 +131,9 @@ def setup_gallery_routes() -> APIRouter:
         file_hash = hashlib.sha256(content).hexdigest()
         db = SessionLocal()
         try:
+            if album_id and user is not None:
+                _get_or_404_album(db, album_id, user)
+
             # SECURITY: scope the dup-detect to THIS user — otherwise a
             # caller can probe whether someone else uploaded the same
             # file (the response leaks the existing row's id+filename).
@@ -69,7 +148,7 @@ def setup_gallery_routes() -> APIRouter:
                 return {"ok": False, "duplicate": True, "filename": existing.filename,
                         "id": existing.id, "message": "Duplicate photo skipped"}
 
-            img_dir = Path("data/generated_images")
+            img_dir = Path(GENERATED_IMAGES_DIR)
             img_dir.mkdir(parents=True, exist_ok=True)
 
             ext = file.filename.rsplit(".", 1)[-1].lower() if "." in file.filename else "png"
@@ -135,7 +214,7 @@ def setup_gallery_routes() -> APIRouter:
                 raise HTTPException(400, "No image provided")
 
             content = await read_upload_limited(file, GALLERY_UPLOAD_MAX_BYTES, "Gallery replacement")
-            img_dir = Path("data/generated_images")
+            img_dir = Path(GENERATED_IMAGES_DIR)
             img_dir.mkdir(parents=True, exist_ok=True)
             img_path = img_dir / _sanitize_gallery_filename(img.filename)
             img_path.write_bytes(content)
@@ -211,7 +290,7 @@ def setup_gallery_routes() -> APIRouter:
             if not user or img.owner != user:
                 raise HTTPException(403, "Not your image")
 
-            img_path = Path("data/generated_images") / img.filename
+            img_path = _gallery_image_path(img.filename)
             if not img_path.exists():
                 raise HTTPException(404, "Image file not found")
 
@@ -248,7 +327,7 @@ def setup_gallery_routes() -> APIRouter:
         """AI upscale using img2img with the diffusion server."""
         import base64, httpx
 
-        require_privilege(request, "can_generate_images")
+        user = require_privilege(request, "can_generate_images")
         form = await request.form()
         file = form.get("image")
         if not file: raise HTTPException(400, "No image")
@@ -260,7 +339,7 @@ def setup_gallery_routes() -> APIRouter:
         # Find image endpoint
         db = SessionLocal()
         try:
-            ep = db.query(ModelEndpoint).filter(ModelEndpoint.model_type == "image", ModelEndpoint.is_enabled == True).first()
+            ep = _first_visible_image_endpoint(db, user)
         finally:
             db.close()
 
@@ -291,7 +370,7 @@ def setup_gallery_routes() -> APIRouter:
         """Style transfer using img2img with the diffusion server."""
         import base64, httpx
 
-        require_privilege(request, "can_generate_images")
+        user = require_privilege(request, "can_generate_images")
         form = await request.form()
         file = form.get("image")
         prompt = form.get("prompt", "")
@@ -303,7 +382,7 @@ def setup_gallery_routes() -> APIRouter:
 
         db = SessionLocal()
         try:
-            ep = db.query(ModelEndpoint).filter(ModelEndpoint.model_type == "image", ModelEndpoint.is_enabled == True).first()
+            ep = _first_visible_image_endpoint(db, user)
         finally:
             db.close()
 
@@ -397,8 +476,7 @@ def setup_gallery_routes() -> APIRouter:
                 .outerjoin(DbSession, GalleryImage.session_id == DbSession.id)
                 .filter(GalleryImage.is_active == True)
             )
-            if user is not None:
-                q = q.filter(GalleryImage.owner == user)
+            q = _owner_filter(q, user)
 
             # Search filter (prompt + tags + ai_tags)
             if search:
@@ -500,23 +578,27 @@ def setup_gallery_routes() -> APIRouter:
         db = SessionLocal()
         try:
             q = db.query(GalleryAlbum)
-            if user:
-                q = q.filter(GalleryAlbum.owner == user)
+            q = _owner_filter(q, user, GalleryAlbum)
             albums = q.order_by(GalleryAlbum.created_at.desc()).all()
             result = []
             for a in albums:
-                count = db.query(GalleryImage).filter(
+                _count_q = db.query(GalleryImage).filter(
                     GalleryImage.album_id == a.id, GalleryImage.is_active == True
-                ).count()
+                )
+                _count_q = _owner_filter(_count_q, user)
+                count = _count_q.count()
                 cover_url = None
                 if a.cover_id:
-                    cover = db.query(GalleryImage).filter(GalleryImage.id == a.cover_id).first()
+                    cover_q = db.query(GalleryImage).filter(GalleryImage.id == a.cover_id)
+                    cover = _owner_filter(cover_q, user).first()
                     if cover:
                         cover_url = f"/api/generated-image/{cover.filename}"
                 elif count > 0:
-                    first = db.query(GalleryImage).filter(
+                    _cover_q = db.query(GalleryImage).filter(
                         GalleryImage.album_id == a.id, GalleryImage.is_active == True
-                    ).order_by(GalleryImage.created_at.desc()).first()
+                    )
+                    _cover_q = _owner_filter(_cover_q, user)
+                    first = _cover_q.order_by(GalleryImage.created_at.desc()).first()
                     if first:
                         cover_url = f"/api/generated-image/{first.filename}"
                 result.append({
@@ -558,10 +640,9 @@ def setup_gallery_routes() -> APIRouter:
             base = db.query(GalleryImage).filter(GalleryImage.is_active == True)
             size_q = db.query(func.sum(GalleryImage.file_size)).filter(GalleryImage.is_active == True)
             album_q = db.query(GalleryAlbum)
-            if user:
-                base = base.filter(GalleryImage.owner == user)
-                size_q = size_q.filter(GalleryImage.owner == user)
-                album_q = album_q.filter(GalleryAlbum.owner == user)
+            base = _owner_filter(base, user)
+            size_q = _owner_filter(size_q, user)
+            album_q = _owner_filter(album_q, user, GalleryAlbum)
             total = base.count()
             total_size = size_q.scalar() or 0
             fav_count = base.filter(GalleryImage.favorite == True).count()
@@ -589,8 +670,7 @@ def setup_gallery_routes() -> APIRouter:
                 GalleryImage.is_active == True,
                 (GalleryImage.ai_tags == None) | (GalleryImage.ai_tags == ""),
             )
-            if user:
-                q = q.filter(GalleryImage.owner == user)
+            q = _owner_filter(q, user)
             if album_id:
                 q = q.filter(GalleryImage.album_id == album_id)
             untagged = q.count()
@@ -649,7 +729,14 @@ def setup_gallery_routes() -> APIRouter:
             if req.favorite is not None:
                 img.favorite = req.favorite
             if req.album_id is not None:
-                img.album_id = req.album_id if req.album_id else None
+                if req.album_id:
+                    # Validate the target album belongs to the caller before
+                    # moving the image into it — mirrors add_to_album, so you
+                    # cannot file your image into another user's album.
+                    _get_or_404_album(db, req.album_id, user)
+                    img.album_id = req.album_id
+                else:
+                    img.album_id = None
             db.commit()
             db.refresh(img)
             return _image_to_dict(img)
@@ -692,11 +779,11 @@ def setup_gallery_routes() -> APIRouter:
             used = set()
             with zipfile.ZipFile(buf, "w", zipfile.ZIP_DEFLATED) as zf:
                 for img in imgs:
-                    src = os.path.join("data", "generated_images", img.filename)
-                    if not os.path.exists(src):
+                    src = _gallery_image_path(img.filename)
+                    if not src.exists():
                         continue
-                    ext = os.path.splitext(img.filename)[1] or ".png"
-                    base = (img.prompt or "").strip() or os.path.splitext(img.filename)[0]
+                    ext = src.suffix or ".png"
+                    base = (img.prompt or "").strip() or src.stem
                     base = re.sub(r"[^\w\-. ]+", "", base)[:60].strip() or img.id
                     name = f"{base}{ext}"
                     i = 1
@@ -818,9 +905,9 @@ def setup_gallery_routes() -> APIRouter:
 
             img_filename = img.filename
             # Remove the file from disk
-            img_path = os.path.join("data", "generated_images", img_filename)
-            if os.path.exists(img_path):
-                os.remove(img_path)
+            img_path = _gallery_image_path(img_filename)
+            if img_path.exists():
+                img_path.unlink()
 
             # Soft-delete the record
             img.is_active = False
@@ -923,7 +1010,7 @@ def setup_gallery_routes() -> APIRouter:
         the request for /v1/images/edits (multipart, inverted mask). Otherwise
         proxy through to a self-hosted diffusion server's /v1/images/inpaint."""
         import httpx
-        require_privilege(request, "can_generate_images")
+        user = require_privilege(request, "can_generate_images")
         body = await request.json()
         # Use endpoint from request body (editor dropdown) or fall back to DB lookup
         base = (body.pop("_endpoint", "") or "").rstrip("/")
@@ -942,14 +1029,11 @@ def setup_gallery_routes() -> APIRouter:
         if not base:
             db = SessionLocal()
             try:
-                eps = db.query(ModelEndpoint).filter(
-                    ModelEndpoint.is_enabled == True,
-                    ModelEndpoint.model_type == "image",
-                ).all()
-                if not eps:
+                ep = _first_visible_image_endpoint(db, user)
+                if not ep:
                     raise HTTPException(400, "No image generation endpoint configured. Serve a diffusion model via Cookbook first.")
-                base = eps[0].base_url.rstrip("/")
-                api_key = eps[0].api_key
+                base = ep.base_url.rstrip("/")
+                api_key = ep.api_key
             finally:
                 db.close()
         else:
@@ -966,10 +1050,12 @@ def setup_gallery_routes() -> APIRouter:
             _target = _norm_url(base)
             db = SessionLocal()
             try:
-                for ep in db.query(ModelEndpoint).all():
-                    if _norm_url(ep.base_url) == _target:
-                        api_key = ep.api_key
-                        break
+                ep = _visible_image_endpoint_for_base(db, _target, user)
+                if ep:
+                    base = (ep.base_url or base).rstrip("/")
+                    api_key = ep.api_key
+                elif user and not _current_user_is_admin(request, user):
+                    raise HTTPException(403, "Choose a registered image endpoint")
             finally:
                 db.close()
 
@@ -1121,7 +1207,7 @@ def setup_gallery_routes() -> APIRouter:
         you get edge blending + lighting unification while keeping the
         composition recognisable."""
         import httpx, base64 as _b64
-        require_privilege(request, "can_generate_images")
+        user = require_privilege(request, "can_generate_images")
         body = await request.json()
 
         image_b64 = body.get("image")
@@ -1148,23 +1234,22 @@ def setup_gallery_routes() -> APIRouter:
         if not base:
             db = SessionLocal()
             try:
-                eps = db.query(ModelEndpoint).filter(
-                    ModelEndpoint.is_enabled == True,
-                    ModelEndpoint.model_type == "image",
-                ).all()
-                if not eps:
+                ep = _first_visible_image_endpoint(db, user)
+                if not ep:
                     raise HTTPException(400, "No image generation endpoint configured.")
-                base = eps[0].base_url.rstrip("/")
-                api_key = eps[0].api_key
+                base = ep.base_url.rstrip("/")
+                api_key = ep.api_key
             finally:
                 db.close()
         else:
             db = SessionLocal()
             try:
-                for ep in db.query(ModelEndpoint).all():
-                    if ep.base_url.rstrip("/").removesuffix("/v1").rstrip("/") == base.rstrip("/").removesuffix("/v1").rstrip("/"):
-                        api_key = ep.api_key
-                        break
+                ep = _visible_image_endpoint_for_base(db, base, user)
+                if ep:
+                    base = (ep.base_url or base).rstrip("/")
+                    api_key = ep.api_key
+                elif user and not _current_user_is_admin(request, user):
+                    raise HTTPException(403, "Choose a registered image endpoint")
             finally:
                 db.close()
 
@@ -1316,6 +1401,7 @@ def setup_gallery_routes() -> APIRouter:
     @router.post("/api/image/sharpen")
     async def sharpen_image(request: Request):
         """Apply unsharp-mask sharpening to an image."""
+        require_privilege(request, "can_generate_images")
         body = await request.json()
         image_b64 = body.get("image")
         amount = body.get("amount", 50) / 100.0
@@ -1635,9 +1721,10 @@ def setup_gallery_routes() -> APIRouter:
         db = SessionLocal()
         try:
             album = _get_or_404_album(db, album_id, user)
-            db.query(GalleryImage).filter(GalleryImage.album_id == album_id).update(
-                {"album_id": None}, synchronize_session=False
-            )
+            q = db.query(GalleryImage).filter(GalleryImage.album_id == album_id)
+            if user is not None:
+                q = q.filter(GalleryImage.owner == user)
+            q.update({"album_id": None}, synchronize_session=False)
             db.delete(album)
             db.commit()
             return {"ok": True}
@@ -1708,7 +1795,7 @@ def setup_gallery_routes() -> APIRouter:
         try:
             img = _get_or_404_image(db, image_id, user)
 
-            img_path = Path("data/generated_images") / img.filename
+            img_path = _gallery_image_path(img.filename)
             if not img_path.exists():
                 raise HTTPException(404, "Image file not found")
 
@@ -1726,7 +1813,7 @@ def setup_gallery_routes() -> APIRouter:
                 return {"error": "Vision is disabled — enable it in Settings → Vision"}
             configured = vl_settings.get("vision_model", "")
             try:
-                chat_url, model_name, headers = _resolve_vl_model(configured)
+                chat_url, model_name, headers = _resolve_vl_model(configured, owner=user)
             except ValueError:
                 return {"error": "No vision model configured — set one in Settings → Vision"}
             if not chat_url:
@@ -1807,4 +1894,3 @@ def setup_gallery_routes() -> APIRouter:
             db.close()
 
     return router
-

routes/history_routes.py

@@ -490,7 +490,13 @@ def setup_history_routes(session_manager) -> APIRouter:
             # Copy messages up to keep_count
             msgs_to_copy = source.history[:keep_count]
             for msg in msgs_to_copy:
-                new_session.add_message(ChatMessage(msg.role, msg.content, msg.metadata))
+                # Copy the metadata dict. Sharing it would let the fork's
+                # persistence (add_message -> _persist_message stamps
+                # _db_id/timestamp onto the dict) mutate the SOURCE session's
+                # in-memory messages, corrupting their _db_id and breaking
+                # edit/delete-by-id on the original conversation.
+                meta = dict(msg.metadata) if isinstance(msg.metadata, dict) else None
+                new_session.add_message(ChatMessage(msg.role, msg.content, meta))
             try:
                 from src.event_bus import fire_event
                 fire_event("session_created", getattr(source, 'owner', None))
@@ -522,6 +528,8 @@ def setup_history_routes(session_manager) -> APIRouter:
     async def compact_session(request: Request, session_id: str):
         """Manually trigger context compaction for a session."""
         _verify_session_owner(request, session_id)
+        from src.auth_helpers import effective_user
+        owner = effective_user(request)
         try:
             session = session_manager.get_session(session_id)
         except KeyError:
@@ -555,7 +563,7 @@ def setup_history_routes(session_manager) -> APIRouter:
             )
 
             # Use utility model if available
-            util_url, util_model, util_headers = resolve_endpoint("utility")
+            util_url, util_model, util_headers = resolve_endpoint("utility", owner=owner or None)
             compact_url = util_url or session.endpoint_url
             compact_model = util_model or session.model
             compact_headers = util_headers if util_url else session.headers

routes/hwfit_routes.py

@@ -1,7 +1,9 @@
 import re
 from copy import deepcopy
 
-from fastapi import APIRouter
+from fastapi import APIRouter, HTTPException
+
+from routes._validators import validate_remote_host, validate_ssh_port
 
 
 # Backends the manual hardware simulator accepts. Must stay a subset of what
@@ -11,6 +13,14 @@ from fastapi import APIRouter
 _MANUAL_BACKENDS = {"cuda", "rocm", "metal", "cpu_x86", "cpu_arm"}
 
 
+def _validate_detection_target(host: str = "", ssh_port: str = "") -> tuple[str, str]:
+    host_value = validate_remote_host(host) or ""
+    port_value = validate_ssh_port(ssh_port) or ""
+    if port_value and not host_value:
+        raise HTTPException(400, "ssh_port requires host")
+    return host_value, port_value
+
+
 def _apply_manual_hardware(system, manual_mode="", manual_gpu_count="", manual_vram_gb="", manual_ram_gb="", manual_backend=""):
     """Manual hardware is a "what if I had this setup" simulator —
     REPLACES the detected hardware entirely instead of adding to it.
@@ -105,6 +115,7 @@ def setup_hwfit_routes():
         """Detect and return current system hardware info. Pass host=user@server for remote.
         fresh=true bypasses the per-host cache (the Rescan button)."""
         from services.hwfit.hardware import detect_system
+        host, ssh_port = _validate_detection_target(host, ssh_port)
         return detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh)
 
     @router.get("/models")
@@ -118,6 +129,7 @@ def setup_hwfit_routes():
         from services.hwfit.hardware import detect_system
         from services.hwfit.fit import rank_models
         from services.hwfit.models import get_models, model_catalog_path
+        host, ssh_port = _validate_detection_target(host, ssh_port)
         system = deepcopy(detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh))
         if system.get("error"):
             return {"system": system, "models": [], "error": system["error"]}
@@ -165,8 +177,14 @@ def setup_hwfit_routes():
             system["gpu_name"] = g["name"]
             system["active_group"] = {**g, "use_count": n}
 
-        if gpu_count != "":
-            n = int(gpu_count)
+        # Parse the optional count defensively (matches the gpu_group guard
+        # above): a non-numeric query param previously raised ValueError ->
+        # HTTP 500. A malformed value is ignored, same as omitting it.
+        try:
+            n = int(gpu_count) if gpu_count != "" else None
+        except ValueError:
+            n = None
+        if n is not None:
             if n == 0:
                 # RAM-only mode: rank against system memory, offload allowed.
                 system["has_gpu"] = False
@@ -196,7 +214,24 @@ def setup_hwfit_routes():
         if target_context is not None:
             target_context = max(1024, min(target_context, 1000000))
 
-        results = rank_models(system, use_case=use_case or None, limit=limit, search=search or None, sort=sort, quant=quant or None, target_context=target_context, fit_only=fit_only)
+        rank_kwargs = {
+            "use_case": use_case or None,
+            "limit": limit,
+            "search": search or None,
+            "sort": sort,
+            "quant": quant or None,
+            "fit_only": fit_only,
+        }
+        if target_context is not None:
+            rank_kwargs["target_context"] = target_context
+        try:
+            import inspect
+            supported = set(inspect.signature(rank_models).parameters)
+            rank_kwargs = {k: v for k, v in rank_kwargs.items() if k in supported}
+        except Exception:
+            rank_kwargs.pop("target_context", None)
+            rank_kwargs.pop("fit_only", None)
+        results = rank_models(system, **rank_kwargs)
         return {"system": system, "models": results}
 
     @router.get("/profiles")
@@ -212,6 +247,7 @@ def setup_hwfit_routes():
         from services.hwfit.hardware import detect_system
         from services.hwfit.models import get_models
         from services.hwfit.profiles import compute_serve_profiles
+        host, ssh_port = _validate_detection_target(host, ssh_port)
         system = detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh)
         if system.get("error"):
             return {"system": system, "profiles": [], "error": system["error"]}
@@ -262,6 +298,7 @@ def setup_hwfit_routes():
         """Rank image generation models against detected hardware."""
         from services.hwfit.hardware import detect_system
         from services.hwfit.image_models import rank_image_models
+        host, ssh_port = _validate_detection_target(host, ssh_port)
         system = deepcopy(detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh))
         if system.get("error"):
             return {"system": system, "models": [], "error": system["error"]}

routes/mcp_routes.py

@@ -13,7 +13,7 @@ import httpx
 
 from core.database import McpServer, SessionLocal
 from core.middleware import require_admin
-from src.constants import DATA_DIR
+from src.constants import DATA_DIR, MCP_OAUTH_DIR
 from src.mcp_manager import McpManager
 
 logger = logging.getLogger(__name__)
@@ -23,7 +23,7 @@ router = APIRouter(prefix="/api/mcp", tags=["mcp"])
 
 def _mcp_oauth_base_dir() -> Path:
     """Directory that may contain OAuth files managed by Odysseus."""
-    return (Path(DATA_DIR) / "mcp_oauth").resolve(strict=False)
+    return Path(MCP_OAUTH_DIR).resolve(strict=False)
 
 
 def _resolve_mcp_oauth_path(raw_path, field_name: str) -> str:

routes/memory_routes.py

@@ -29,11 +29,10 @@ from src.llm_core import llm_call_async
 from services.memory.memory_extractor import audit_memories
 from src.auth_helpers import get_current_user, require_user
 from src.endpoint_resolver import resolve_endpoint
-from src.upload_limits import read_upload_limited
+from src.upload_limits import read_upload_limited, MEMORY_IMPORT_MAX_BYTES
 
 logger = logging.getLogger(__name__)
 
-MEMORY_IMPORT_MAX_BYTES = int(os.getenv("ODYSSEUS_MEMORY_IMPORT_MAX_BYTES", str(10 * 1024 * 1024)))
 
 def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionManager, memory_vector=None):
     """Set up memory-related routes."""
@@ -106,6 +105,13 @@ def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionM
         if memory_manager.find_duplicates(text, user_mem):
             return {"ok": True, "count": len(user_mem), "message": "Memory already exists"}
 
+        if memory_data.session_id:
+            try:
+                session_obj = session_manager.get_session(memory_data.session_id)
+            except KeyError:
+                raise HTTPException(404, "Session not found")
+            _assert_session_owner(session_obj, user)
+
         new_entry = memory_manager.add_entry(text, memory_data.source, memory_data.category, owner=user)
         if memory_data.session_id:
             new_entry["session_id"] = memory_data.session_id
@@ -164,8 +170,17 @@ def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionM
 
             session_id = memory.get("session_id")
             if session_id and session_id in session_manager.sessions:
-                session = session_manager.get_session(session_id)
-                memory["session_name"] = session.name if session else f"Session {session_id[:6]}"
+                try:
+                    session = session_manager.get_session(session_id)
+                    if session:
+                        _assert_session_owner(session, user)
+                    memory["session_name"] = session.name if session else f"Session {session_id[:6]}"
+                except KeyError:
+                    memory["session_name"] = "Unknown"
+                except HTTPException as exc:
+                    if exc.status_code != 404:
+                        raise
+                    memory["session_name"] = "Unknown"
             else:
                 memory["session_name"] = "Unknown"
 
@@ -371,7 +386,7 @@ def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionM
                 tmp.write(content)
                 tmp_path = tmp.name
             try:
-                text = _process_pdf(tmp_path)
+                text = _process_pdf(tmp_path, owner=_owner(request))
             finally:
                 os.unlink(tmp_path)
         else:

routes/model_routes.py

@@ -4,6 +4,7 @@ import os
 import re
 import uuid
 import json
+import hashlib
 import socket
 import time as _time
 import logging
@@ -122,6 +123,21 @@ def _clear_user_pref_endpoint_refs(all_prefs: dict, ep_id: str) -> int:
     return cleared_users
 
 
+def _default_endpoint_needs_assignment(current_default_id: str, enabled_endpoint_ids) -> bool:
+    """Whether the global default chat endpoint should be (re)assigned.
+
+    True when nothing is configured yet, or the configured default no longer
+    resolves to an enabled endpoint (e.g. the user disabled it). Without the
+    second case, adding a new endpoint after disabling the previous default
+    leaves `default_endpoint_id` pointing at the disabled endpoint, so features
+    that read the raw setting (Memory → Tidy) fail with "No default model
+    configured" even though an enabled endpoint exists. See #3586.
+    """
+    if not current_default_id:
+        return True
+    return current_default_id not in enabled_endpoint_ids
+
+
 # Loopback hosts a user might type for a local model server (LM Studio,
 # llama.cpp, vLLM, …). Inside Docker these point at the *container*, not the
 # host the server actually runs on.
@@ -282,6 +298,7 @@ _HOST_TO_CURATED = (
     ("fireworks.ai", "fireworks"),
     ("googleapis.com", "google"),
     ("x.ai", "xai"),
+    ("nvidia.com", "nvidia"),
     ("openrouter.ai", "openrouter"),
     ("ollama.com", "ollama"),
 )
@@ -476,10 +493,17 @@ _NON_CHAT_PREFIXES = (
     "dall-e", "tts-", "whisper", "text-embedding", "embedding",
     "davinci", "babbage", "moderation", "omni-moderation",
     "sora", "gpt-image", "chatgpt-image",
+    # embedding / retrieval / non-chat models (common across providers)
+    "snowflake/arctic-embed", "nvidia/nv-embed", "embed",
 )
 _NON_CHAT_CONTAINS = (
     "-realtime", "-transcribe", "-tts", "-codex",
-    "codex-",
+    "codex-", "content-safety", "-safety", "-reward", "nvclip",
+    "kosmos", "fuyu", "deplot", "vila", "neva",
+    "gliner", "riva", "-parse", "-embedqa", "-nemoretriever",
+    "topic-control", "calibration",
+    "ai-synthetic-video", "cosmos-reason2",
+    "bge", "llama-guard",
 )
 _NON_CHAT_EXACT_PREFIXES = (
     "gpt-audio",  # gpt-audio, gpt-audio-mini etc. (not gpt-4o-audio-preview which is chat)
@@ -502,9 +526,71 @@ def _is_chat_model(model_id: str) -> bool:
     return True
 
 
+def _delete_orphaned_provider_auth(db, auth_id: Optional[str], exclude_ep_id: Optional[str] = None) -> bool:
+    """Delete a ProviderAuthSession once no endpoint still references it."""
+    if not auth_id:
+        return False
+    from core.database import ProviderAuthSession
+    still_referenced = db.query(ModelEndpoint.id).filter(
+        ModelEndpoint.provider_auth_id == auth_id,
+        ModelEndpoint.id != exclude_ep_id,
+    ).first()
+    if still_referenced is not None:
+        return False
+    auth_row = db.query(ProviderAuthSession).filter(ProviderAuthSession.id == auth_id).first()
+    if auth_row is None:
+        return False
+    db.delete(auth_row)
+    return True
+
+
+def _safe_detect_provider(base_url: str) -> str:
+    """Best-effort provider detection that must not break endpoint probing."""
+    try:
+        return _detect_provider(base_url)
+    except Exception as exc:
+        logger.debug("Provider detection failed for %s: %s", base_url, exc)
+        return ""
+
+
+def _safe_build_models_url(base_url: str) -> str:
+    """Build a /models URL without letting optional provider imports break probes."""
+    try:
+        return build_models_url(base_url)
+    except Exception as exc:
+        logger.debug("Model URL detection failed for %s: %s", base_url, exc)
+        return f"{(base_url or '').rstrip('/')}/models"
+
+
+def _safe_build_headers(api_key: Optional[str], base_url: str) -> dict:
+    """Build auth headers without letting optional provider imports break probes."""
+    try:
+        return build_headers(api_key, base_url)
+    except Exception as exc:
+        logger.debug("Header detection failed for %s: %s", base_url, exc)
+        return {"Authorization": f"Bearer {api_key}"} if api_key else {}
+
+
+def _is_discovery_only_provider(provider: str) -> bool:
+    return provider == "chatgpt-subscription"
+
+
+def _resolve_probe_key(ep) -> Optional[str]:
+    """API key/bearer to probe an endpoint with."""
+    try:
+        from src.endpoint_resolver import resolve_endpoint_runtime
+        _base, key = resolve_endpoint_runtime(ep, owner=getattr(ep, "owner", None))
+        return key
+    except Exception as exc:
+        logger.warning("Probe key resolution failed for %s: %s", getattr(ep, "id", "?"), exc)
+        return None
+
+
 def _probe_single_model(base: str, api_key: str, model_id: str, timeout: int = 10, with_tools: bool = False) -> dict:
     """Send a realistic completion request to a single model. Returns {status, latency_ms, error?}."""
-    provider = _detect_provider(base)
+    provider = _safe_detect_provider(base)
+    if _is_discovery_only_provider(provider):
+        return {"status": "ok", "latency_ms": 0, "skipped": True}
     messages = [
         {"role": "system", "content": "You are a helpful assistant."},
         {"role": "user", "content": "Say OK"},
@@ -523,12 +609,12 @@ def _probe_single_model(base: str, api_key: str, model_id: str, timeout: int = 1
     elif provider == "ollama":
         from src.llm_core import _build_ollama_payload
         target_url = build_chat_url(base)
-        h = build_headers(api_key, base)
+        h = _safe_build_headers(api_key, base)
         h["Content-Type"] = "application/json"
         payload = _build_ollama_payload(model_id, messages, 0.0, 5, stream=False, tools=_test_tools)
     else:
         target_url = build_chat_url(base)
-        h = build_headers(api_key, base)
+        h = _safe_build_headers(api_key, base)
         h["Content-Type"] = "application/json"
         from src.llm_core import _uses_max_completion_tokens, _restricts_temperature
         _max_key = "max_completion_tokens" if _uses_max_completion_tokens(model_id) else "max_tokens"
@@ -618,9 +704,15 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
     For Anthropic, queries their /v1/models API, falling back to hardcoded list."""
     from src.endpoint_resolver import resolve_url
     base = resolve_url(_normalize_base(base_url))
-    if _detect_provider(base) == "anthropic":
+    provider = _safe_detect_provider(base)
+    if provider == "chatgpt-subscription":
+        from src.chatgpt_subscription import fetch_available_models
+        if api_key:
+            return fetch_available_models(api_key, timeout=timeout)
+        return []
+    if provider == "anthropic":
         # Try Anthropic's /v1/models endpoint first
-        url = build_models_url(base)
+        url = _safe_build_models_url(base)
         headers = {"anthropic-version": "2023-06-01"}
         if api_key:
             headers["x-api-key"] = api_key
@@ -643,8 +735,8 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
                 return []
             logger.warning(f"Anthropic /v1/models failed, using hardcoded list: {e}")
         return list(ANTHROPIC_MODELS)
-    url = build_models_url(base)
-    headers = build_headers(api_key, base)
+    url = _safe_build_models_url(base)
+    headers = _safe_build_headers(api_key, base)
     try:
         r = httpx.get(url, headers=headers, timeout=timeout, verify=llm_verify())
         r.raise_for_status()
@@ -662,7 +754,7 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
                 for _e in _PROVIDER_CURATED.get(_ck, []):
                     if _e not in set(models) and not any(m.startswith(_e) for m in models):
                         models.append(_e)
-            return models
+            return [m for m in models if _is_chat_model(m)]
     except httpx.HTTPStatusError as e:
         if api_key:
             status = e.response.status_code if e.response is not None else "unknown"
@@ -686,7 +778,7 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
             data = r.json()
             models = [m.get("name") or m.get("model") for m in (data.get("models") or []) if m.get("name") or m.get("model")]
             if models:
-                return models
+                return [m for m in models if _is_chat_model(m)]
     except Exception as e:
         logger.debug(f"Ollama /api/tags probe failed for {base}: {e}")
     # Fall back to curated list if the provider has a URL-based match (e.g. z.ai has no /models endpoint)
@@ -702,7 +794,7 @@ def _ping_endpoint(base_url: str, api_key: str = None, timeout: float = 1.5) ->
     """Reachability probe that does not require installed/listed models."""
     from src.endpoint_resolver import resolve_url
     base = resolve_url(_normalize_base(base_url))
-    headers = build_headers(api_key, base)
+    headers = _safe_build_headers(api_key, base)
 
     # Ollama exposes /v1/models (OpenAI-compatible) AND native /api/version,
     # /api/tags. Probe native paths for Ollama-style endpoints, but avoid using
@@ -754,7 +846,22 @@ def _ping_endpoint(base_url: str, api_key: str = None, timeout: float = 1.5) ->
 
     try:
         r = httpx.get(base, headers=headers, timeout=timeout, verify=llm_verify())
-        return _result_from_response(r)
+        result = _result_from_response(r)
+        if result["reachable"]:
+            return result
+        sc = result.get("status_code") or 0
+        if 400 <= sc < 500 and sc not in (401, 403):
+            models_url = _safe_build_models_url(base)
+            try:
+                r2 = httpx.get(models_url, headers=headers, timeout=timeout, verify=llm_verify())
+                result2 = _result_from_response(r2)
+                if result2["reachable"]:
+                    return result2
+            except Exception:
+                pass
+        if sc:
+            return result
+        last_error = result.get("error") or last_error
     except Exception as e:
         last_error = str(e)[:120]
 
@@ -850,6 +957,14 @@ def _visible_models(cached_models, hidden_models, pinned_models=None):
     return [m for m in merged if m not in hidden]
 
 
+def _api_key_fingerprint(api_key: Optional[str]) -> str:
+    """Stable, non-secret label for distinguishing same-URL credentials."""
+    key = (api_key or "").strip()
+    if not key:
+        return ""
+    return hashlib.sha256(key.encode("utf-8")).hexdigest()[:8]
+
+
 def setup_model_routes(model_discovery):
     router = APIRouter(prefix="/api")
 
@@ -1028,7 +1143,7 @@ def setup_model_routes(model_discovery):
 
         for ep in endpoints:
             base = _normalize_base(ep.base_url)
-            provider = _detect_provider(base)
+            provider = _safe_detect_provider(base)
             # Merge cached + pinned models, then filter out hidden ones
             ep_model_type = getattr(ep, "model_type", None) or "llm"
             model_ids = _visible_models(
@@ -1104,8 +1219,9 @@ def setup_model_routes(model_discovery):
                 raise HTTPException(401, "Not authenticated")
         except HTTPException:
             raise
-        except Exception:
-            pass
+        except Exception as e:
+            logger.error("Auth gate error in GET /api/models, failing closed: %s", e)
+            raise HTTPException(status_code=500, detail="Internal error")
         # Admins see every endpoint (they manage the global pool); regular
         # users get the owner-scoped view.
         _is_admin = False
@@ -1169,7 +1285,14 @@ def setup_model_routes(model_discovery):
             t0 = _time.time()
             try:
                 import asyncio as _asyncio
-                ping = await _asyncio.to_thread(_ping_endpoint, data["base"], data.get("api_key"), 1.5)
+                # Bumped 1.5s → 3.5s. The previous 1.5s budget was clipping
+                # local vLLM endpoints on Tailscale links where the model
+                # server is still loading (Qwen3.5-122B takes 2–3 min to
+                # warm); /v1/models can take 500–2500 ms on a busy box,
+                # which pushed _ping_endpoint's full path-discovery sweep
+                # past the cap and marked the row offline despite the
+                # user actively chatting with it.
+                ping = await _asyncio.to_thread(_ping_endpoint, data["base"], data.get("api_key"), 3.5)
                 lat = round((_time.time() - t0) * 1000)
                 return {
                     "alive": bool(ping.get("reachable")),
@@ -1207,7 +1330,7 @@ def setup_model_routes(model_discovery):
         results = []
         for ep in endpoints:
             base = _normalize_base(ep.base_url)
-            provider = _detect_provider(base)
+            provider = _safe_detect_provider(base)
             kind = _effective_endpoint_kind(ep, base)
             cached_count = len(_cached_model_ids(ep))
             entry = {
@@ -1386,10 +1509,35 @@ def setup_model_routes(model_discovery):
                 # admin-pinned IDs that a probe would never surface.
                 status = "online" if (all_models or pinned) else "offline"
                 ping = None
+                # When cached_models is empty, do a quick reachability probe.
+                # Bumped 1.0s → 3.5s because the user reported endpoints they
+                # were ACTIVELY chatting with showed "offline" — the previous
+                # 1s timeout was clipping live cloud endpoints (DeepSeek can
+                # take 1.5–2.5s on /v1/models when their region is under load,
+                # vLLM on a remote GPU box behind SSH can also push past 1s).
+                # 3.5s still keeps the picker render snappy in the common
+                # "everything's already cached" path because this branch only
+                # runs for endpoints with an empty cached_models.
                 if not all_models and not pinned and r.is_enabled:
-                    ping = _ping_endpoint(r.base_url, r.api_key, timeout=1.0)
+                    ping = _ping_endpoint(r.base_url, r.api_key, timeout=3.5)
                     if ping.get("reachable"):
                         status = "empty"
+                        # Best-effort: if the probe came back reachable, try
+                        # to populate cached_models in the background so the
+                        # NEXT picker load shows "online" instead of "empty".
+                        # Failure here is silent — we already returned the
+                        # "empty" status, and the existing background refresh
+                        # path will eventually fill it in too.
+                        try:
+                            probed = _probe_endpoint(r.base_url, r.api_key, timeout=5)
+                            if probed:
+                                r.cached_models = json.dumps(probed)
+                                db.commit()
+                                all_models = probed
+                                visible = _visible_models(all_models, r.hidden_models, pinned)
+                                status = "online"
+                        except Exception as _refill_err:
+                            logger.debug(f"opportunistic cached_models refill failed for {r.id}: {_refill_err!r}")
                 base = _normalize_base(r.base_url)
                 kind = _effective_endpoint_kind(r, base)
                 results.append({
@@ -1397,6 +1545,7 @@ def setup_model_routes(model_discovery):
                     "name": r.name,
                     "base_url": r.base_url,
                     "has_key": bool(r.api_key),
+                    "api_key_fingerprint": _api_key_fingerprint(r.api_key),
                     "is_enabled": r.is_enabled,
                     "models": visible,
                     "pinned_models": pinned,
@@ -1469,15 +1618,27 @@ def setup_model_routes(model_discovery):
         # re-adding manually-added endpoints under their host:port name.
         from src.auth_helpers import get_current_user as _gcu_dedup
         _caller = _gcu_dedup(request) or None
+        _incoming_api_key = api_key.strip()
         _db_dedup = SessionLocal()
         try:
-            existing = (
+            _same_url_rows = (
                 _db_dedup.query(ModelEndpoint)
                 .filter(ModelEndpoint.base_url == base_url)
                 .filter((ModelEndpoint.owner.is_(None)) | (ModelEndpoint.owner == _caller))
                 .order_by(ModelEndpoint.owner.desc())  # prefer owned over shared
-                .first()
+                .all()
             )
+            existing = None
+            _empty_key_existing = None
+            for _candidate in _same_url_rows:
+                _candidate_key = (getattr(_candidate, "api_key", None) or "").strip()
+                if _candidate_key == _incoming_api_key:
+                    existing = _candidate
+                    break
+                if _incoming_api_key and not _candidate_key and _empty_key_existing is None:
+                    _empty_key_existing = _candidate
+            if existing is None and _incoming_api_key and _empty_key_existing is not None:
+                existing = _empty_key_existing
             if existing:
                 changed = False
                 # Persist any incoming pinned IDs onto the existing row. An
@@ -1526,6 +1687,8 @@ def setup_model_routes(model_discovery):
                     "id": existing.id,
                     "name": existing.name,
                     "base_url": existing.base_url,
+                    "has_key": bool(existing.api_key),
+                    "api_key_fingerprint": _api_key_fingerprint(existing.api_key),
                     "models": _visible_models(
                         existing_models,
                         getattr(existing, "hidden_models", None),
@@ -1579,12 +1742,19 @@ def setup_model_routes(model_discovery):
             )
             db.add(ep)
             db.commit()
-            # Auto-set as default chat endpoint if none configured yet. Seed
-            # the first CHAT model (not raw model_ids[0]) so we don't pin the
-            # global default to an embedding/tts/etc. entry a provider happens
-            # to list first.
+            # Auto-set as default chat endpoint when none is usable yet — either
+            # nothing is configured, or the configured default points at an
+            # endpoint that is now missing/disabled (#3586). Seed the first CHAT
+            # model (not raw model_ids[0]) so we don't pin the global default to
+            # an embedding/tts/etc. entry a provider happens to list first.
             settings = _load_settings()
-            if not settings.get("default_endpoint_id"):
+            enabled_ids = {
+                e.id
+                for e in db.query(ModelEndpoint).filter(
+                    ModelEndpoint.is_enabled == True  # noqa: E712
+                ).all()
+            }
+            if _default_endpoint_needs_assignment(settings.get("default_endpoint_id") or "", enabled_ids):
                 from src.endpoint_resolver import _first_chat_model
                 settings["default_endpoint_id"] = ep.id
                 settings["default_model"] = _first_chat_model(model_ids) or ""
@@ -1599,6 +1769,8 @@ def setup_model_routes(model_discovery):
             "id": ep_id,
             "name": name.strip(),
             "base_url": base_url,
+            "has_key": bool(api_key.strip()),
+            "api_key_fingerprint": _api_key_fingerprint(api_key),
             "models": _merge_model_ids(model_ids, _pinned),
             "pinned_models": _pinned,
             "online": bool(model_ids) or bool(_pinned) or bool(ping.get("reachable")),
@@ -2049,7 +2221,9 @@ def setup_model_routes(model_discovery):
             cleared_user_preferences = _clear_user_prefs_for_endpoint(ep_id)
             cleared_sessions = _clear_sessions_for_endpoint(db, ep.base_url)
             cleared_loaded_sessions = _clear_loaded_sessions_for_endpoint(ep.base_url)
+            auth_id = getattr(ep, "provider_auth_id", None)
             db.delete(ep)
+            cleared_provider_auth = _delete_orphaned_provider_auth(db, auth_id, exclude_ep_id=ep_id)
             db.commit()
             _invalidate_models_cache()
             _local_probe_cache["data"] = None
@@ -2059,6 +2233,7 @@ def setup_model_routes(model_discovery):
                 "cleared_user_preferences": cleared_user_preferences,
                 "cleared_sessions": cleared_sessions,
                 "cleared_loaded_sessions": cleared_loaded_sessions,
+                "cleared_provider_auth": cleared_provider_auth,
             }
         finally:
             db.close()

routes/note_routes.py

@@ -11,6 +11,7 @@ from pydantic import BaseModel
 
 from core.database import SessionLocal, Note
 from src.auth_helpers import get_current_user
+from src.constants import DATA_DIR
 from sqlalchemy.orm.attributes import flag_modified
 
 logger = logging.getLogger(__name__)
@@ -95,6 +96,32 @@ def _note_to_dict(note: Note) -> Dict[str, Any]:
     }
 
 
+def _reminder_text_from_note(note: Note) -> tuple[str, str]:
+    """Return the reminder title/body from a stored note row."""
+    title = (note.title or "Note reminder").strip() or "Note reminder"
+    if note.items:
+        try:
+            items = json.loads(note.items)
+        except (json.JSONDecodeError, TypeError):
+            items = None
+        if isinstance(items, list):
+            pending: list[str] = []
+            for item in items:
+                if not isinstance(item, dict):
+                    continue
+                if item.get("done") or item.get("checked"):
+                    continue
+                text = str(item.get("text") or "").strip()
+                if text:
+                    pending.append(text)
+            if pending:
+                shown = "\n".join(f"- {text}" for text in pending[:8])
+                extra = f"\n...and {len(pending) - 8} more" if len(pending) > 8 else ""
+                return title, f"Pending ({len(pending)}):\n{shown}{extra}"
+            return title, f"{len(items)} item{'s' if len(items) != 1 else ''}"
+    return title, (note.content or "").strip()[:400]
+
+
 
 # ---------------------------------------------------------------------------
 # Reminder dispatch — module-level so background tasks (built-in actions)
@@ -114,8 +141,9 @@ async def dispatch_reminder(
     note_id: str,
     owner: str = "",
     queue_browser: bool = True,
+    settings_override: dict | None = None,
 ) -> dict:
-    """Fire a reminder via the configured channel (browser/email/ntfy).
+    """Fire a reminder via the configured channel (browser/email/ntfy/webhook).
 
     Args:
         title: short headline shown to the user
@@ -129,7 +157,7 @@ async def dispatch_reminder(
     nothing is "sent" synchronously for it — the channel just routes there.
     """
     from src.settings import load_settings
-    settings = load_settings()
+    settings = {**load_settings(), **(settings_override or {})}
     channel = settings.get("reminder_channel", "browser")
     llm_on = bool(settings.get("reminder_llm_synthesis", False))
     title = (title or "").strip()
@@ -143,7 +171,7 @@ async def dispatch_reminder(
             from datetime import datetime as _dt, timezone as _tz, timedelta as _td
             from pathlib import Path as _P
             _slug = "".join(c if (c.isalnum() or c in "-_.@") else "_" for c in (owner or "default"))
-            cache_path = _P(f"data/note_pings_{_slug}.json")
+            cache_path = _P(DATA_DIR) / f"note_pings_{_slug}.json"
             if cache_path.exists():
                 cache = _json.loads(cache_path.read_text(encoding="utf-8"))
             last = cache.get(cache_key)
@@ -160,13 +188,14 @@ async def dispatch_reminder(
                 # Treat those as browser-only dedupe so email reminders can be
                 # retried by the backend scanner after a failed frontend path.
                 should_skip = last_dt >= _dt.now(_tz.utc) - _td(minutes=25)
-                if should_skip and channel in ("email", "ntfy"):
+                if should_skip and channel in ("email", "ntfy", "webhook"):
                     should_skip = last_channel == channel
                 if should_skip:
                     return {
                         "synthesis": None,
                         "email_sent": False,
                         "ntfy_sent": False,
+                        "webhook_sent": False,
                         "browser_sent": True,
                         "skipped": True,
                     }
@@ -179,9 +208,9 @@ async def dispatch_reminder(
         try:
             from src.endpoint_resolver import resolve_endpoint
             from src.llm_core import llm_call_async
-            url, model, headers = resolve_endpoint("utility")
+            url, model, headers = resolve_endpoint("utility", owner=owner or None)
             if not url:
-                url, model, headers = resolve_endpoint("default")
+                url, model, headers = resolve_endpoint("default", owner=owner or None)
             if url and model:
                 raw = await llm_call_async(
                     url=url, model=model,
@@ -360,6 +389,76 @@ async def dispatch_reminder(
             email_error = str(e) or e.__class__.__name__
             logger.warning(f"Reminder email send failed: {e}")
 
+    webhook_sent = False
+    webhook_error = ""
+    if channel == "webhook":
+        try:
+            import httpx
+            import json as _wjson
+            from src.integrations import load_integrations
+            # Built-in payload defaults for known presets so users don't have
+            # to configure a template just to use a standard service.
+            _PRESET_TEMPLATE_DEFAULTS = {
+                "discord_webhook": '{"embeds": [{"title": "{{title}}", "description": "{{message}}", "color": 5793266}]}',
+            }
+            intg_id = settings.get("reminder_webhook_integration_id", "").strip()
+            template = settings.get("reminder_webhook_payload_template", "").strip()
+            if not intg_id:
+                webhook_error = "No webhook integration selected"
+            else:
+                intg = next(
+                    (i for i in load_integrations()
+                     if i.get("id") == intg_id and i.get("base_url")),
+                    None,
+                )
+                if not intg:
+                    webhook_error = f"Integration {intg_id!r} not found or missing base URL"
+                else:
+                    # Fall back to a built-in default for known presets so
+                    # users don't have to configure a template for standard
+                    # services like Discord.
+                    if not template:
+                        template = _PRESET_TEMPLATE_DEFAULTS.get(intg.get("preset", ""), "")
+                    if not template:
+                        webhook_error = "No payload template configured"
+                    else:
+                        # Render template: JSON-escape the values so the result
+                        # is always valid JSON regardless of special characters.
+                        # dumps() returns `"value"` — strip outer quotes.
+                        msg = (synthesis or note_body or title or "Reminder")[:4000]
+                        _t = _wjson.dumps(title or "Reminder")[1:-1]
+                        _m = _wjson.dumps(msg)[1:-1]
+                        rendered = template.replace("{{title}}", _t).replace("{{message}}", _m)
+                        hdrs = {"Content-Type": "application/json"}
+                        api_key = intg.get("api_key", "")
+                        auth_type = (intg.get("auth_type") or "none").lower()
+                        if api_key:
+                            if auth_type == "bearer":
+                                hdrs["Authorization"] = f"Bearer {api_key}"
+                            elif auth_type == "header":
+                                hdrs[intg.get("auth_header") or "Authorization"] = api_key
+                        url = intg["base_url"].rstrip("/")
+                        # SSRF guard — matches the pattern used by webhook_routes,
+                        # CalDAV, search, and embeddings. Blocks link-local / metadata
+                        # addresses (169.254.x.x) by default; set
+                        # REMINDER_WEBHOOK_BLOCK_PRIVATE_IPS=true to also block
+                        # RFC-1918 ranges for locked-down deployments.
+                        import os as _os
+                        from src.url_safety import check_outbound_url as _chk
+                        _block = _os.getenv("REMINDER_WEBHOOK_BLOCK_PRIVATE_IPS", "false").lower() == "true"
+                        _ok, _reason = _chk(url, block_private=_block)
+                        if not _ok:
+                            webhook_error = f"Webhook URL rejected: {_reason}"
+                        else:
+                            async with httpx.AsyncClient(timeout=10.0) as client:
+                                resp = await client.post(url, content=rendered.encode(), headers=hdrs)
+                                webhook_sent = resp.is_success
+                                if not webhook_sent:
+                                    webhook_error = f"Webhook returned HTTP {resp.status_code}"
+        except Exception as e:
+            webhook_error = str(e) or e.__class__.__name__
+            logger.warning(f"Reminder webhook send failed: {e}")
+
     ntfy_sent = False
     ntfy_error = ""
     if channel == "ntfy":
@@ -415,7 +514,7 @@ async def dispatch_reminder(
     # second send for the same note within 25 min. Without this, a note
     # whose due_date fires while the user has the app open got TWO emails
     # (frontend-fired here + background-fired by ping_notes 0–5 min later).
-    if (email_sent or ntfy_sent or browser_sent or local_browser_sent) and note_id:
+    if (email_sent or ntfy_sent or webhook_sent or browser_sent or local_browser_sent) and note_id:
         try:
             import json as _json
             from datetime import datetime as _dt, timezone as _tz
@@ -425,13 +524,13 @@ async def dispatch_reminder(
             _STATE = cache_path
             if _STATE is None:
                 _slug = "".join(c if (c.isalnum() or c in "-_.@") else "_" for c in (owner or "default"))
-                _STATE = _P(f"data/note_pings_{_slug}.json")
+                _STATE = _P(DATA_DIR) / f"note_pings_{_slug}.json"
             _STATE.parent.mkdir(parents=True, exist_ok=True)
             try:
                 _cache = cache or (_json.loads(_STATE.read_text(encoding="utf-8")) if _STATE.exists() else {})
             except Exception:
                 _cache = {}
-            sent_channel = "email" if email_sent else "ntfy" if ntfy_sent else "browser"
+            sent_channel = "email" if email_sent else "ntfy" if ntfy_sent else "webhook" if webhook_sent else "browser"
             _cache[cache_key or str(note_id)] = {
                 "at": _dt.now(_tz.utc).isoformat(),
                 "channel": sent_channel,
@@ -441,11 +540,14 @@ async def dispatch_reminder(
             logger.debug(f"dispatch_reminder: cache write failed: {_e}")
 
     return {
+        "channel": channel,
         "synthesis": synthesis,
         "email_sent": email_sent,
         "email_error": email_error,
         "ntfy_sent": ntfy_sent,
         "ntfy_error": ntfy_error,
+        "webhook_sent": webhook_sent,
+        "webhook_error": webhook_error,
         "browser_sent": browser_sent or local_browser_sent,
     }
 
@@ -467,6 +569,23 @@ def setup_note_routes(task_scheduler=None):
     def _owner(request: Request) -> Optional[str]:
         return get_current_user(request)
 
+    def _is_admin_or_single_user(request: Request, user: str | None) -> bool:
+        if user == "internal-tool":
+            return True
+        if not user:
+            # require_user() already admitted this request, which only happens
+            # for auth-disabled, loopback-bypass, or unconfigured single-user
+            # modes. There is no separate non-admin account boundary there.
+            return True
+        try:
+            from core.auth import AuthManager
+            auth_mgr = getattr(request.app.state, "auth_manager", None) or AuthManager()
+            if not getattr(auth_mgr, "is_configured", True):
+                return True
+            return bool(auth_mgr.is_admin(user))
+        except Exception:
+            return False
+
     # --- LIST ---
     @router.get("")
     def list_notes(
@@ -684,20 +803,46 @@ def setup_note_routes(task_scheduler=None):
         """
         # Gate against anonymous callers — LLM synthesis can burn tokens.
         from src.auth_helpers import require_user as _ru
-        _ru(request)
+        user = _ru(request)
         body = await request.json()
-        note_id = body.get("note_id")
-        title = (body.get("title") or "").strip()
-        note_body = (body.get("body") or "").strip()
+        note_id = str(body.get("note_id") or "").strip()
         if not note_id:
             raise HTTPException(400, "note_id required")
 
-        # Delegate to the module-level helper so background tasks can reuse
-        # the same dispatch without an HTTP roundtrip + auth cookie.
+        caller = _owner(request)
+        is_test = note_id.startswith("test-")
+        is_admin = _is_admin_or_single_user(request, user or caller)
+        _override: dict = {}
+        if is_test:
+            if not is_admin:
+                raise HTTPException(403, "Admin only")
+            title = (body.get("title") or "Test Reminder").strip() or "Test Reminder"
+            note_body = (body.get("body") or "").strip()
+            # Optional overrides let the admin settings test button pass the
+            # current UI values directly so it never races a pending save.
+            if body.get("channel"):
+                _override["reminder_channel"] = body["channel"]
+            if body.get("webhook_integration_id"):
+                _override["reminder_webhook_integration_id"] = body["webhook_integration_id"]
+            if body.get("webhook_payload_template"):
+                _override["reminder_webhook_payload_template"] = body["webhook_payload_template"]
+        else:
+            db = SessionLocal()
+            try:
+                note = db.query(Note).filter(Note.id == note_id).first()
+                if not note:
+                    raise HTTPException(404, "Note not found")
+                if caller is not None and note.owner != caller:
+                    raise HTTPException(404, "Note not found")
+                title, note_body = _reminder_text_from_note(note)
+            finally:
+                db.close()
+
         return await dispatch_reminder(
             title=title, note_body=note_body, note_id=note_id,
-            owner=_owner(request) or "",
+            owner=caller or "",
             queue_browser=False,
+            settings_override=_override or None,
         )
 
     # --- REORDER NOTES ---

routes/personal_routes.py

@@ -6,16 +6,14 @@ import uuid
 from typing import List, Tuple
 from fastapi import APIRouter, HTTPException, Query, Request, UploadFile, File, Depends
 from src.request_models import DirectoryRequest
-from core.constants import BASE_DIR, PERSONAL_DIR
+from core.constants import BASE_DIR, PERSONAL_DIR, PERSONAL_UPLOADS_DIR
 from src.rag_singleton import get_rag_manager
-from src.auth_helpers import get_current_user, require_user
+from src.auth_helpers import require_privilege, require_user
 from core.middleware import require_admin
 from src.upload_handler import secure_filename
+from src.upload_limits import PERSONAL_UPLOAD_MAX_BYTES
 
-UPLOADS_DIR = os.path.join(BASE_DIR, "data", "personal_uploads")
-MAX_PERSONAL_UPLOAD_BYTES = int(
-    os.getenv("ODYSSEUS_PERSONAL_UPLOAD_MAX_BYTES", str(25 * 1024 * 1024))
-)
+UPLOADS_DIR = PERSONAL_UPLOADS_DIR
 
 logger = logging.getLogger(__name__)
 
@@ -194,7 +192,7 @@ def setup_personal_routes(personal_docs_manager, rag_manager, rag_available):
     @router.post("/upload")
     async def upload_files_to_rag(request: Request, files: List[UploadFile] = File(...)):
         """Upload files directly into RAG. Supports text and PDF."""
-        user = get_current_user(request)
+        user = require_privilege(request, "can_use_documents")
         rag = _rag()
         if not rag:
             raise HTTPException(503, "RAG system is not available — is the embedding service running?")
@@ -208,8 +206,8 @@ def setup_personal_routes(personal_docs_manager, rag_manager, rag_available):
         for upload in files:
             try:
                 file_path, stored_name, safe_name = _unique_personal_upload_path(upload_dir, upload.filename)
-                content_bytes = await upload.read(MAX_PERSONAL_UPLOAD_BYTES + 1)
-                if len(content_bytes) > MAX_PERSONAL_UPLOAD_BYTES:
+                content_bytes = await upload.read(PERSONAL_UPLOAD_MAX_BYTES + 1)
+                if len(content_bytes) > PERSONAL_UPLOAD_MAX_BYTES:
                     logger.warning(f"Rejected oversized personal upload: {upload.filename!r}")
                     total_failed += 1
                     continue
@@ -286,9 +284,12 @@ def setup_personal_routes(personal_docs_manager, rag_manager, rag_available):
             except ValueError:
                 # commonpath raises on mixed drives / non-comparable paths
                 in_uploads = False
-            if in_uploads and abs_target != base_abs and os.path.exists(abs_target):
-                os.remove(abs_target)
-                deleted_from_disk = True
+            if in_uploads and abs_target != base_abs:
+                try:
+                    os.remove(abs_target)
+                    deleted_from_disk = True
+                except FileNotFoundError:
+                    pass  # already gone — race with another request or cleanup
 
             # Exclude the file from the listing (persists across restarts)
             personal_docs_manager.exclude_file(filepath)

routes/prefs_routes.py

@@ -4,8 +4,9 @@ import os
 from typing import Optional
 from fastapi import APIRouter, Request
 from src.auth_helpers import get_current_user
+from src.constants import USER_PREFS_FILE
 
-PREFS_FILE = os.path.join("data", "user_prefs.json")
+PREFS_FILE = USER_PREFS_FILE
 
 
 def _load():

routes/preset_routes.py

@@ -9,6 +9,7 @@ from pydantic import BaseModel, Field
 
 from src.request_models import PresetUpdateRequest
 from core.middleware import require_admin
+from src.auth_helpers import effective_user
 
 logger = logging.getLogger(__name__)
 
@@ -100,7 +101,8 @@ def setup_preset_routes(preset_manager) -> APIRouter:
 
         try:
             model_spec = data.get("model") or ""
-            url, model, headers = _resolve_model(model_spec)
+            user = effective_user(request)
+            url, model, headers = _resolve_model(model_spec, owner=user)
             result = await llm_call_async(url, model, messages, temperature=0.8, max_tokens=500, headers=headers)
             return {"success": True, "prompt": result.strip()}
         except Exception as e:

routes/research_routes.py

@@ -14,6 +14,7 @@ from fastapi.responses import HTMLResponse, StreamingResponse
 from pydantic import BaseModel, Field
 from src.endpoint_resolver import resolve_endpoint
 from src.auth_helpers import _auth_disabled, get_current_user
+from src.constants import DEEP_RESEARCH_DIR
 
 _SESSION_ID_RE = re.compile(r"^[a-zA-Z0-9-]{1,128}$")
 
@@ -37,13 +38,15 @@ def _first_chat_model(models) -> str:
     return (models[0] if models else "")
 
 
-def _resolve_research_endpoint(sess) -> tuple:
+def _resolve_research_endpoint(sess, owner: Optional[str] = None) -> tuple:
     """Return (endpoint_url, model, headers) for Deep Research, checking admin overrides."""
+    owner = owner or getattr(sess, "owner", None) or None
     url, model, headers = resolve_endpoint(
         "research",
         fallback_url=sess.endpoint_url,
         fallback_model=sess.model,
         fallback_headers=sess.headers,
+        owner=owner,
     )
     return url, model, headers
 
@@ -72,6 +75,38 @@ def _owned_enabled_endpoint(db, owner, endpoint_id=None):
     return owner_filter(q, ModelEndpoint, owner).first()
 
 
+def _resolve_endpoint_runtime(ep, owner=None, model: Optional[str] = None):
+    """Resolve a ModelEndpoint row into (chat_url, model, headers).
+
+    Mirrors endpoint_resolver.resolve_endpoint's provider-auth handling for
+    panel-selected research endpoints. ChatGPT Subscription endpoints keep
+    OAuth tokens in ProviderAuthSession, so ep.api_key is intentionally empty.
+    """
+    from src.endpoint_resolver import (
+        build_chat_url,
+        build_headers,
+        resolve_endpoint_runtime as resolve_model_endpoint_runtime,
+    )
+
+    try:
+        base, api_key = resolve_model_endpoint_runtime(ep, owner=owner)
+    except Exception as e:
+        logger.warning("Could not resolve endpoint credentials for research: %s", e)
+        return None
+
+    ep_model = (model or "").strip()
+    if not ep_model:
+        try:
+            models = json.loads(ep.cached_models) if ep.cached_models else []
+            if models:
+                ep_model = _first_chat_model(models)
+        except Exception:
+            pass
+    if not ep_model:
+        return None
+    return build_chat_url(base), ep_model, build_headers(api_key, base)
+
+
 def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
     router = APIRouter(tags=["research"])
 
@@ -98,7 +133,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
         if entry is not None:
             return entry.get("owner", "") == user
         # Task no longer in memory — check the persisted JSON.
-        path = Path("data/deep_research") / f"{session_id}.json"
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
         if not path.exists():
             return False
         try:
@@ -162,7 +197,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
     def _assert_owns_research(session_id: str, user: str) -> None:
         """404-not-403 ownership gate for a research session's on-disk JSON.
         Use BEFORE returning any data or mutating the file."""
-        path = Path("data/deep_research") / f"{session_id}.json"
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
         if not path.exists():
             raise HTTPException(404, "Research not found")
         try:
@@ -225,7 +260,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
     ):
         user = _require_user(request)
         """List all completed research for the Library panel."""
-        data_dir = Path("data/deep_research")
+        data_dir = Path(DEEP_RESEARCH_DIR)
         items = []
         for p in data_dir.glob("*.json"):
             try:
@@ -275,7 +310,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
         summary, stats — used by the Library preview panel."""
         user = _require_user(request)
         _validate_session_id(session_id)
-        path = Path("data/deep_research") / f"{session_id}.json"
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
         if not path.exists():
             raise HTTPException(404, "Research not found")
         try:
@@ -292,7 +327,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
         """Soft-archive / restore a research report (sets `archived` in its JSON)."""
         user = _require_user(request)
         _validate_session_id(session_id)
-        path = Path("data/deep_research") / f"{session_id}.json"
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
         if not path.exists():
             raise HTTPException(404, "Research not found")
         try:
@@ -312,7 +347,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
         """Delete a research result from disk."""
         user = _require_user(request)
         _validate_session_id(session_id)
-        data_dir = Path("data/deep_research")
+        data_dir = Path(DEEP_RESEARCH_DIR)
         json_path = data_dir / f"{session_id}.json"
         deleted = False
         if json_path.exists():
@@ -368,7 +403,6 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
 
         if body.endpoint_id:
             from src.database import SessionLocal
-            from src.endpoint_resolver import normalize_base, build_chat_url, build_headers
             db = SessionLocal()
             try:
                 # Owner-scoped: never resolve another user's private endpoint
@@ -377,35 +411,26 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
                 ep = _owned_enabled_endpoint(db, user, body.endpoint_id)
                 if not ep:
                     raise HTTPException(404, "Endpoint not found or disabled")
-                base = normalize_base(ep.base_url)
-                ep_url = build_chat_url(base)
-                ep_headers = build_headers(ep.api_key, base)
-                ep_model = body.model or ""
-                if not ep_model:
-                    try:
-                        import json as _json
-                        models = _json.loads(ep.cached_models) if ep.cached_models else []
-                        if models:
-                            ep_model = _first_chat_model(models)
-                    except Exception:
-                        pass
+                resolved = _resolve_endpoint_runtime(ep, owner=user, model=body.model)
+                if not resolved:
+                    raise HTTPException(400, "Endpoint is not configured with a usable model.")
+                ep_url, ep_model, ep_headers = resolved
             finally:
                 db.close()
         else:
-            ep_url, ep_model, ep_headers = resolve_endpoint("research")
+            ep_url, ep_model, ep_headers = resolve_endpoint("research", owner=user)
             if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("utility")
+                ep_url, ep_model, ep_headers = resolve_endpoint("utility", owner=user)
             # When neither research nor utility is configured, use the user's
             # configured DEFAULT model (default_endpoint_id/default_model) rather
             # than arbitrarily grabbing the first enabled endpoint's first model
             # (which surfaced gpt-3.5). "Default" should mean the default model.
             if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("default")
+                ep_url, ep_model, ep_headers = resolve_endpoint("default", owner=user)
             if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("chat")
+                ep_url, ep_model, ep_headers = resolve_endpoint("chat", owner=user)
             if not ep_url:
                 from src.database import SessionLocal
-                from src.endpoint_resolver import normalize_base, build_chat_url, build_headers
                 db = SessionLocal()
                 try:
                     # Owner-scoped first-enabled fallback: the caller's own rows
@@ -414,18 +439,9 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
                     # /api/v1/chat fallback (webhook_routes._first_enabled_endpoint).
                     ep = _owned_enabled_endpoint(db, user)
                     if ep:
-                        base = normalize_base(ep.base_url)
-                        ep_url = build_chat_url(base)
-                        ep_headers = build_headers(ep.api_key, base)
-                        ep_model = ""
-                        if ep.cached_models:
-                            try:
-                                import json as _json
-                                models = _json.loads(ep.cached_models)
-                                if models:
-                                    ep_model = _first_chat_model(models)
-                            except Exception:
-                                pass
+                        resolved = _resolve_endpoint_runtime(ep, owner=user)
+                        if resolved:
+                            ep_url, ep_model, ep_headers = resolved
                 finally:
                     db.close()
             if not ep_url:
@@ -494,7 +510,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
             raise HTTPException(404, "No research found for this session")
         result = research_handler.get_result(session_id)
         if result is None:
-            p = Path("data/deep_research") / f"{session_id}.json"
+            p = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
             if p.exists():
                 d = json.loads(p.read_text(encoding="utf-8"))
                 return {
@@ -534,7 +550,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
         sources = research_handler.get_sources(session_id) or []
         query = ""
 
-        path = Path("data/deep_research") / f"{session_id}.json"
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
         if path.exists():
             try:
                 disk = json.loads(path.read_text(encoding="utf-8"))
@@ -572,19 +588,18 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
                 ep_headers = dict(r_headers)
 
         if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("chat"))
+            _merge(*resolve_endpoint("chat", owner=user))
         if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("research"))
+            _merge(*resolve_endpoint("research", owner=user))
         if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("utility"))
+            _merge(*resolve_endpoint("utility", owner=user))
         if not ep_url or not ep_model:
-            # Last resort: any enabled endpoint
+            # Last resort: this user's enabled endpoint, plus legacy shared rows.
             from src.database import SessionLocal
-            from src.database import ModelEndpoint
             from src.endpoint_resolver import normalize_base, build_chat_url, build_headers
             db = SessionLocal()
             try:
-                ep = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True).first()
+                ep = _owned_enabled_endpoint(db, user)
                 if ep:
                     base = normalize_base(ep.base_url)
                     fallback_url = build_chat_url(base)
@@ -594,7 +609,7 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
                         try:
                             models = json.loads(ep.cached_models)
                             if models:
-                                fallback_model = models[0]
+                                fallback_model = _first_chat_model(models)
                         except Exception:
                             pass
                     _merge(fallback_url, fallback_model, fallback_headers)

routes/session_routes.py

@@ -10,8 +10,9 @@ import logging
 from core.session_manager import SessionManager
 from core.models import ChatMessage
 from src.request_models import SessionResponse
-from core.database import Session as DbSession, SessionLocal, Document, GalleryImage
-from src.auth_helpers import get_current_user, effective_user
+from core.database import Session as DbSession, SessionLocal, Document, GalleryImage, utcnow_naive
+from src.auth_helpers import get_current_user, effective_user, _auth_disabled, owner_filter
+from src.session_actions import is_session_recently_active
 
 
 def _sanitize_export_filename(name: str) -> str:
@@ -92,35 +93,30 @@ def _reject_compact_during_active_run(session_id: str) -> None:
 
 
 def _verify_session_owner(request: Request, session_id: str, session_manager=None):
-    """Verify the current user owns the session. Raises 404 if not.
+    """Verify the current user owns the session, honoring single-user modes.
 
-    Ownership is checked against the DB row when one exists (unchanged). If
-    there is no DB row but the caller owns an in-memory "ghost" session — one
-    that lives only in ``session_manager`` because it was never persisted, or
-    its DB row was removed out-of-band — fall back to the in-memory owner so the
-    user can still manage and delete it. Without this fallback such sessions are
-    listed by ``/api/sessions`` (they come from the in-memory manager) yet every
-    per-session operation 404s, making them impossible to delete (issue #1044).
-
-    ``session_manager`` is optional and defaults to ``None`` so existing callers
-    that only care about persisted sessions keep their exact prior behavior.
+    Authenticated requests must match the stored DB or in-memory owner. When
+    auth is disabled and no user is present, treat the app as single-user mode:
+    verify that the session exists, but do not compare its stored owner. This
+    keeps QA/dev instances with AUTH_ENABLED=false from rejecting owner-stamped
+    rows created while auth was previously enabled.
     """
     user = effective_user(request)
-    if not user:
-        raise HTTPException(403, "Authentication required")
+    if not user and not _auth_disabled():
+        raise HTTPException(401, "Authentication required")
     db = SessionLocal()
     try:
         row = db.query(DbSession.owner).filter(DbSession.id == session_id).first()
     finally:
         db.close()
     if row is not None:
-        if row.owner != user:
+        if user and row.owner != user:
             raise HTTPException(404, f"Session {session_id} not found")
         return
     # No DB row — allow the caller to act on an in-memory ghost they own.
     if session_manager is not None:
         ghost = getattr(session_manager, "sessions", {}).get(session_id)
-        if ghost is not None and getattr(ghost, "owner", None) == user:
+        if ghost is not None and (not user or getattr(ghost, "owner", None) == user):
             return
     raise HTTPException(404, f"Session {session_id} not found")
 
@@ -262,7 +258,9 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             last_msg_map = {}
             mode_map = {}
             msg_count_map = {}
-            rows = db.query(DbSession.id, DbSession.folder, DbSession.total_input_tokens, DbSession.total_output_tokens, DbSession.is_important, DbSession.created_at, DbSession.updated_at, DbSession.last_message_at, DbSession.mode, DbSession.message_count).filter(DbSession.archived == False).all()
+            q = db.query(DbSession.id, DbSession.folder, DbSession.total_input_tokens, DbSession.total_output_tokens, DbSession.is_important, DbSession.created_at, DbSession.updated_at, DbSession.last_message_at, DbSession.mode, DbSession.message_count).filter(DbSession.archived == False)
+            q = owner_filter(q, DbSession, user)
+            rows = q.all()
             for row in rows:
                 folder_map[row.id] = row.folder
                 token_map[row.id] = (row.total_input_tokens or 0) + (row.total_output_tokens or 0)
@@ -281,15 +279,19 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             # Sessions with active documents that have content
             from sqlalchemy import func
             doc_session_ids = set(
-                r[0] for r in db.query(Document.session_id)
-                .filter(Document.is_active == True,
-                        Document.current_content != None,
-                        func.trim(Document.current_content) != "")
+                r[0] for r in owner_filter(
+                    db.query(Document.session_id)
+                    .filter(Document.is_active == True,
+                            Document.current_content != None,
+                            func.trim(Document.current_content) != ""),
+                    Document, user)
                 .distinct().all()
             )
             img_session_ids = set(
-                r[0] for r in db.query(GalleryImage.session_id)
-                .filter(GalleryImage.session_id != None)
+                r[0] for r in owner_filter(
+                    db.query(GalleryImage.session_id)
+                    .filter(GalleryImage.session_id != None),
+                    GalleryImage, user)
                 .distinct().all()
             )
         finally:
@@ -370,8 +372,13 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             pass
         elif not model_to_use:
             from src.llm_core import list_model_ids
-            ids = list_model_ids(endpoint_url, timeout=REQUEST_TIMEOUT,
-                                 headers=validation_headers)
+            ids = list_model_ids(
+                endpoint_url,
+                timeout=REQUEST_TIMEOUT,
+                headers=validation_headers,
+                owner=user,
+                endpoint_id=endpoint_id.strip() if endpoint_id else None,
+            )
             if not ids:
                 raise HTTPException(400, "Cannot reach /v1/models")
             # Default to the first CHAT model — endpoints often list embedding/
@@ -385,8 +392,13 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             from src.llm_core import list_model_ids
             import os as _os
             req_base = _os.path.basename(model_to_use.rstrip("/"))
-            avail = list_model_ids(endpoint_url, timeout=REQUEST_TIMEOUT,
-                                   headers=validation_headers)
+            avail = list_model_ids(
+                endpoint_url,
+                timeout=REQUEST_TIMEOUT,
+                headers=validation_headers,
+                owner=user,
+                endpoint_id=endpoint_id.strip() if endpoint_id else None,
+            )
             if not avail:
                 raise HTTPException(400, "Cannot reach /v1/models")
             if model_to_use not in avail:
@@ -543,22 +555,25 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             ids = body.get("ids", [])
         except Exception:
             ids = []
+        deleted_count = 0
         for sid in ids:
             try:
                 _verify_session_owner(request, sid, session_manager)
-                session_manager.delete_session(sid)
+                
+                # Enforce "starred" protection consistent with single-session delete
                 db = SessionLocal()
                 try:
-                    db.query(_CM).filter(_CM.session_id == sid).delete()
-                    db.query(DbSession).filter(DbSession.id == sid).delete()
-                    db.commit()
-                except Exception:
-                    db.rollback()
+                    db_sess = db.query(DbSession).filter(DbSession.id == sid).first()
+                    if db_sess and db_sess.is_important:
+                        continue
                 finally:
                     db.close()
+
+                if session_manager.delete_session(sid):
+                    deleted_count += 1
             except Exception:
                 pass
-        return {"deleted": len(ids)}
+        return {"deleted": deleted_count}
 
     @router.delete("/session/{sid}")
     def delete_session(request: Request, sid: str):
@@ -924,7 +939,8 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
         from src.endpoint_resolver import resolve_endpoint
         from src.llm_core import llm_call_async
 
-        url, model, headers = resolve_endpoint("utility", owner=get_current_user(request))
+        owner = getattr(session, "owner", None) or effective_user(request)
+        url, model, headers = resolve_endpoint("utility", owner=owner)
         if not url or not model:
             url, model, headers = session.endpoint_url, session.model, session.headers
         if not url or not model:
@@ -1006,7 +1022,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
         }
         _THROWAWAY_MAX_MESSAGES = 4  # only delete if <= this many messages
         try:
-            rows = db.query(DbSession).filter(DbSession.archived == False, DbSession.owner == user).all()
+            rows = db.query(DbSession).filter(DbSession.archived == False, DbSession.owner == user).limit(2000).all()
             folder_map = {r.id: r.folder for r in rows}
             # Precompute per-session message counts in TWO aggregate queries
             # instead of 1–3 queries PER session — with many chats the per-row
@@ -1017,6 +1033,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                 db.query(DbMsg.session_id, _sa_func.count(DbMsg.id))
                 .filter(DbMsg.role == "assistant").group_by(DbMsg.session_id).all()
             )
+            cleanup_now = utcnow_naive()
             for row in rows:
                 # Never delete important sessions
                 if getattr(row, 'is_important', False):
@@ -1029,6 +1046,8 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                     if hasattr(session_manager, 'delete_session'):
                         session_manager.delete_session(row.id)
                     continue
+                if is_session_recently_active(row, now=cleanup_now):
+                    continue
                 msg_count = _counts.get(row.id, 0)
                 should_delete = False
                 if msg_count == 0:

routes/shell_routes.py

@@ -13,6 +13,7 @@ import tempfile
 from collections import namedtuple
 from pathlib import Path
 from typing import Dict, Any
+from core.platform_compat import IS_APPLE_SILICON, which_tool
 
 # POSIX-only: `pty`/`fcntl` transitively import `termios`, which does NOT exist
 # on Windows, so importing them unconditionally crashed app startup there
@@ -37,6 +38,7 @@ from core.platform_compat import (
     IS_WINDOWS,
     detached_popen_kwargs,
     find_bash,
+    git_bash_path,
 )
 
 
@@ -92,6 +94,7 @@ def _venv_activate_prefix(venv: str | None) -> str:
     act = venv if venv.endswith("/bin/activate") else venv.rstrip("/") + "/bin/activate"
     return f". {act} && "
 
+
 logger = logging.getLogger(__name__)
 
 PTY_SUPPORTED = pty is not None and fcntl is not None and hasattr(os, "setsid")
@@ -169,7 +172,10 @@ def _package_installed_from_probe(name: str, probe: dict) -> bool:
             and (dists.get("torch") or modules.get("torch", {}).get("real_module"))
         )
     if name == "hf_transfer":
-        return bool(dists.get("hf-transfer") or modules.get("hf_transfer", {}).get("real_module"))
+        return bool(
+            dists.get("hf-transfer")
+            or modules.get("hf_transfer", {}).get("real_module")
+        )
     return bool(dists.get(name) or modules.get(name, {}).get("real_module"))
 
 
@@ -194,8 +200,14 @@ def _package_status_note(name: str, probe: dict) -> str:
         if binaries.get("llama-server"):
             parts.append(f"native llama-server: {binaries['llama-server']}")
         if dists.get("llama-cpp-python"):
-            parts.append(f"python package: llama-cpp-python {dists['llama-cpp-python']}")
-        return "; ".join(parts) if parts else "No native llama-server or llama-cpp-python server package found."
+            parts.append(
+                f"python package: llama-cpp-python {dists['llama-cpp-python']}"
+            )
+        return (
+            "; ".join(parts)
+            if parts
+            else "No native llama-server or llama-cpp-python server package found."
+        )
     if name == "diffusers":
         if _package_installed_from_probe(name, probe):
             return f"diffusers {dists.get('diffusers', 'available')} with torch {dists.get('torch', 'available')}"
@@ -205,7 +217,9 @@ def _package_status_note(name: str, probe: dict) -> str:
     return ""
 
 
-def _package_pip_update_status(pkg: dict, probe: dict | None = None) -> PackageUpdateStatus:
+def _package_pip_update_status(
+    pkg: dict, probe: dict | None = None
+) -> PackageUpdateStatus:
     """Return whether the Dependencies UI should offer a generic pip update.
 
     "Installed" means Cookbook can use the dependency. It does not always mean
@@ -213,12 +227,28 @@ def _package_pip_update_status(pkg: dict, probe: dict | None = None) -> PackageU
     native llama-server can come from a package manager/source build, and a CLI
     may be on PATH without matching Python package metadata.
     """
+    if pkg.get("name") == "APFEL":
+        return PackageUpdateStatus(
+            False,
+            "",  # Note is empty because IT DOES allow for updates outside of PIP.
+        )
+
     if pkg.get("kind") == "system" or not pkg.get("pip"):
-        return PackageUpdateStatus(False, "Update this system dependency outside Odysseus.")
+        return PackageUpdateStatus(
+            False, "Update this system dependency outside Odysseus."
+        )
 
     name = pkg.get("name")
-    binaries = probe.get("binaries") if isinstance(probe, dict) and isinstance(probe.get("binaries"), dict) else {}
-    dists = probe.get("dists") if isinstance(probe, dict) and isinstance(probe.get("dists"), dict) else {}
+    binaries = (
+        probe.get("binaries")
+        if isinstance(probe, dict) and isinstance(probe.get("binaries"), dict)
+        else {}
+    )
+    dists = (
+        probe.get("dists")
+        if isinstance(probe, dict) and isinstance(probe.get("dists"), dict)
+        else {}
+    )
 
     if name == "llama_cpp" and binaries.get("llama-server"):
         return PackageUpdateStatus(
@@ -231,7 +261,9 @@ def _package_pip_update_status(pkg: dict, probe: dict | None = None) -> PackageU
             "Using a vLLM CLI on PATH without Python package metadata; update it outside Odysseus.",
         )
 
-    return PackageUpdateStatus(True, "Update uses pip in the selected Python environment.")
+    return PackageUpdateStatus(
+        True, "Update uses pip in the selected Python environment."
+    )
 
 
 def _prepend_user_install_bins_to_path() -> None:
@@ -250,7 +282,9 @@ def _prepend_user_install_bins_to_path() -> None:
         candidates = []
     candidates.append(os.path.expanduser("~/.local/bin"))
 
-    parts = os.environ.get("PATH", "").split(os.pathsep) if os.environ.get("PATH") else []
+    parts = (
+        os.environ.get("PATH", "").split(os.pathsep) if os.environ.get("PATH") else []
+    )
     changed = False
     for path in reversed([p for p in candidates if p]):
         if path not in parts:
@@ -357,9 +391,11 @@ PTY_UNSUPPORTED_ERROR = "pty_unsupported"
 
 class ShellExecRequest(BaseModel):
     command: str
-    timeout: int | None = None  # optional override; 0 = no timeout (run until client disconnects)
-    use_pty: bool = False       # use pseudo-TTY (for progress bars)
-    use_tmux: bool = False      # run in tmux session (survives browser disconnect)
+    timeout: int | None = (
+        None  # optional override; 0 = no timeout (run until client disconnects)
+    )
+    use_pty: bool = False  # use pseudo-TTY (for progress bars)
+    use_tmux: bool = False  # run in tmux session (survives browser disconnect)
 
 
 async def _create_shell(command: str, **kwargs):
@@ -368,8 +404,16 @@ async def _create_shell(command: str, **kwargs):
     POSIX: /bin/sh via create_subprocess_shell (unchanged behaviour).
     Windows: prefer a real bash (Git Bash/WSL) so bash-syntax commands behave
     the same as on Linux; fall back to cmd.exe when no bash is installed.
+    Powershell commands are executed directly via cmd.exe /c to avoid quoting
+    and env variable expansion errors under Git Bash.
     """
     if IS_WINDOWS:
+        # PowerShell commands (used by the frontend for Windows log-file polling
+        # and session management) must run directly — passing them through
+        # bash -c mangles $env:VAR syntax and breaks the command.
+        cmd_trim = command.strip()
+        if cmd_trim.startswith("powershell") or cmd_trim.startswith("cmd "):
+            return await asyncio.create_subprocess_shell(command, **kwargs)
         bash = find_bash()
         if bash:
             return await asyncio.create_subprocess_exec(bash, "-c", command, **kwargs)
@@ -386,9 +430,7 @@ async def _exec_shell(command: str, timeout: int = EXEC_TIMEOUT) -> Dict[str, An
             stderr=asyncio.subprocess.PIPE,
             cwd=str(Path.home()),
         )
-        stdout_b, stderr_b = await asyncio.wait_for(
-            proc.communicate(), timeout=timeout
-        )
+        stdout_b, stderr_b = await asyncio.wait_for(proc.communicate(), timeout=timeout)
         stdout = stdout_b.decode(errors="replace")[:MAX_OUTPUT]
         stderr = stderr_b.decode(errors="replace")[:MAX_OUTPUT]
         return {"stdout": stdout, "stderr": stderr, "exit_code": proc.returncode}
@@ -399,7 +441,11 @@ async def _exec_shell(command: str, timeout: int = EXEC_TIMEOUT) -> Dict[str, An
                 await proc.wait()
             except ProcessLookupError:
                 pass
-        return {"stdout": "", "stderr": f"Command timed out after {timeout}s", "exit_code": -1}
+        return {
+            "stdout": "",
+            "stderr": f"Command timed out after {timeout}s",
+            "exit_code": -1,
+        }
     except Exception as e:
         return {"stdout": "", "stderr": str(e), "exit_code": -1}
 
@@ -481,7 +527,7 @@ async def _generate_pty(cmd: str, timeout: int, request: Request):
                 if idx == -1:
                     break
                 line = buf[:idx].decode(errors="replace")
-                buf = buf[idx + sep_len:]
+                buf = buf[idx + sep_len :]
                 if line:
                     yield f"data: {json.dumps({'stream': 'stdout', 'data': line})}\n\n"
 
@@ -503,7 +549,7 @@ async def _generate_pty(cmd: str, timeout: int, request: Request):
                 if idx == -1:
                     break
                 line = buf[:idx].decode(errors="replace")
-                buf = buf[idx + sep_len:]
+                buf = buf[idx + sep_len :]
                 if line:
                     yield f"data: {json.dumps({'stream': 'stdout', 'data': line})}\n\n"
             if buf:
@@ -534,6 +580,7 @@ def _pty_read(fd: int) -> bytes | None:
     """Blocking read from PTY fd. Called via run_in_executor.
     Returns bytes on data, None on timeout (no data yet)."""
     import select
+
     r, _, _ = select.select([fd], [], [], 1.0)
     if r:
         try:
@@ -557,10 +604,10 @@ async def _generate_tmux(cmd: str, request: Request):
     script_path = TMUX_LOG_DIR / f"{session_id}.sh"
     script_path.write_text(
         f"#!/bin/bash\n"
-        f"ODYSSEUS_USER_SHELL=\"${{SHELL:-}}\"\n"
-        f"if [ -n \"$ODYSSEUS_USER_SHELL\" ] && [ -x \"$ODYSSEUS_USER_SHELL\" ]; then\n"
-        f"  ODYSSEUS_USER_PATH=\"$(\"$ODYSSEUS_USER_SHELL\" -ic 'printf \"__ODYSSEUS_PATH__%s\\n\" \"$PATH\"' 2>/dev/null | sed -n 's/^__ODYSSEUS_PATH__//p' | tail -n 1 || true)\"\n"
-        f"  if [ -n \"$ODYSSEUS_USER_PATH\" ]; then export PATH=\"$ODYSSEUS_USER_PATH:$PATH\"; fi\n"
+        f'ODYSSEUS_USER_SHELL="${{SHELL:-}}"\n'
+        f'if [ -n "$ODYSSEUS_USER_SHELL" ] && [ -x "$ODYSSEUS_USER_SHELL" ]; then\n'
+        f'  ODYSSEUS_USER_PATH="$("$ODYSSEUS_USER_SHELL" -ic \'printf "__ODYSSEUS_PATH__%s\\n" "$PATH"\' 2>/dev/null | sed -n \'s/^__ODYSSEUS_PATH__//p\' | tail -n 1 || true)"\n'
+        f'  if [ -n "$ODYSSEUS_USER_PATH" ]; then export PATH="$ODYSSEUS_USER_PATH:$PATH"; fi\n'
         f"fi\n"
         f"{cmd} 2>&1 | tee '{log_path}'\n"
         f"EC=${{PIPESTATUS[0]}}\n"
@@ -570,7 +617,9 @@ async def _generate_tmux(cmd: str, request: Request):
         encoding="utf-8",
     )
     script_path.chmod(0o755)
-    logger.info("tmux wrapper script created: session=%s path=%s", session_id, script_path)
+    logger.info(
+        "tmux wrapper script created: session=%s path=%s", session_id, script_path
+    )
 
     tmux_cmd = f"tmux new-session -d -s {session_id} {shlex.quote(str(script_path))}"
 
@@ -602,7 +651,9 @@ async def _generate_tmux(cmd: str, request: Request):
         # Read new lines from log
         try:
             if log_path.exists():
-                lines = log_path.read_text(encoding="utf-8", errors="replace").splitlines()
+                lines = log_path.read_text(
+                    encoding="utf-8", errors="replace"
+                ).splitlines()
                 new_lines = lines[lines_sent:]
                 for line in new_lines:
                     if line.startswith(":::EXIT_CODE:::"):
@@ -630,7 +681,9 @@ async def _generate_tmux(cmd: str, request: Request):
             # Session ended — do one final read
             await asyncio.sleep(0.5)
             if log_path.exists():
-                lines = log_path.read_text(encoding="utf-8", errors="replace").splitlines()
+                lines = log_path.read_text(
+                    encoding="utf-8", errors="replace"
+                ).splitlines()
                 for line in lines[lines_sent:]:
                     if line.startswith(":::EXIT_CODE:::"):
                         try:
@@ -672,8 +725,8 @@ async def _generate_win_detached(cmd: str, request: Request):
     if bash:
         script_path = TMUX_LOG_DIR / f"{session_id}.sh"
         script_path.write_text(
-            f"{cmd} > {shlex.quote(str(log_path))} 2>&1\n"
-            f"echo $? > {shlex.quote(str(exit_path))}\n",
+            f"{cmd} > {shlex.quote(git_bash_path(log_path))} 2>&1\n"
+            f"echo $? > {shlex.quote(git_bash_path(exit_path))}\n",
             encoding="utf-8",
         )
         argv = [bash, str(script_path)]
@@ -711,7 +764,9 @@ async def _generate_win_detached(cmd: str, request: Request):
             return
         try:
             if log_path.exists():
-                lines = log_path.read_text(encoding="utf-8", errors="replace").splitlines()
+                lines = log_path.read_text(
+                    encoding="utf-8", errors="replace"
+                ).splitlines()
                 for line in lines[lines_sent:]:
                     yield f"data: {json.dumps({'stream': 'stdout', 'data': line})}\n\n"
                 lines_sent = len(lines)
@@ -723,11 +778,18 @@ async def _generate_win_detached(cmd: str, request: Request):
             await asyncio.sleep(0.3)
             try:
                 if log_path.exists():
-                    lines = log_path.read_text(encoding="utf-8", errors="replace").splitlines()
+                    lines = log_path.read_text(
+                        encoding="utf-8", errors="replace"
+                    ).splitlines()
                     for line in lines[lines_sent:]:
                         yield f"data: {json.dumps({'stream': 'stdout', 'data': line})}\n\n"
                     lines_sent = len(lines)
-                exit_code = int((exit_path.read_text(encoding="utf-8", errors="replace").strip() or "0"))
+                exit_code = int(
+                    (
+                        exit_path.read_text(encoding="utf-8", errors="replace").strip()
+                        or "0"
+                    )
+                )
             except Exception:
                 exit_code = 0
             break
@@ -753,7 +815,9 @@ def setup_shell_routes() -> APIRouter:
             return {"stdout": "", "stderr": "No command provided", "exit_code": 1}
 
         logger.info("User shell exec requested: length=%d", len(cmd))
-        result = await _exec_shell(cmd, timeout=EXEC_TIMEOUT)
+        result = await _exec_shell(
+            cmd, timeout=req.timeout if req.timeout is not None else EXEC_TIMEOUT
+        )
         return result
 
     @router.post("/api/shell/stream")
@@ -762,9 +826,11 @@ def setup_shell_routes() -> APIRouter:
         _require_admin(request)
         cmd = req.command.strip()
         if not cmd:
+
             async def empty():
                 yield f"data: {json.dumps({'stream': 'stderr', 'data': 'No command provided'})}\n\n"
                 yield f"data: {json.dumps({'exit_code': 1})}\n\n"
+
             return StreamingResponse(empty(), media_type="text/event-stream")
 
         timeout = req.timeout if req.timeout is not None else STREAM_TIMEOUT
@@ -781,7 +847,11 @@ def setup_shell_routes() -> APIRouter:
         if use_tmux:
             # tmux is POSIX-only; Windows uses a detached-process + logfile tail
             # that preserves the "survives disconnect" behaviour.
-            gen = _generate_win_detached(cmd, request) if IS_WINDOWS else _generate_tmux(cmd, request)
+            gen = (
+                _generate_win_detached(cmd, request)
+                if IS_WINDOWS
+                else _generate_tmux(cmd, request)
+            )
             return StreamingResponse(gen, media_type="text/event-stream")
 
         if use_pty and not IS_WINDOWS:
@@ -813,7 +883,12 @@ def setup_shell_routes() -> APIRouter:
                             chunk = await stream.read(4096)
                             if not chunk:
                                 if buf:
-                                    await q.put((name, buf.decode(errors="replace").rstrip("\r\n")))
+                                    await q.put(
+                                        (
+                                            name,
+                                            buf.decode(errors="replace").rstrip("\r\n"),
+                                        )
+                                    )
                                 break
                             buf += chunk
                             while True:
@@ -821,7 +896,7 @@ def setup_shell_routes() -> APIRouter:
                                 if idx == -1:
                                     break
                                 line = buf[:idx].decode(errors="replace")
-                                buf = buf[idx + sep_len:]
+                                buf = buf[idx + sep_len :]
                                 if line:
                                     await q.put((name, line))
                     finally:
@@ -880,7 +955,12 @@ def setup_shell_routes() -> APIRouter:
         return StreamingResponse(generate(), media_type="text/event-stream")
 
     @router.get("/api/cookbook/packages")
-    async def list_packages(request: Request, host: str | None = None, ssh_port: str | None = None, venv: str | None = None):
+    async def list_packages(
+        request: Request,
+        host: str | None = None,
+        ssh_port: str | None = None,
+        venv: str | None = None,
+    ):
         """Check which optional packages are installed.
 
         Local-target packages are checked in-process. Remote-target packages
@@ -890,7 +970,13 @@ def setup_shell_routes() -> APIRouter:
         """
         _require_admin(request)
         _reject_cross_site(request)
-        import importlib, importlib.metadata as importlib_metadata, shlex, json as _json, site, sys
+        import importlib
+        import importlib.metadata as importlib_metadata
+        import shlex
+        import json as _json
+        import site
+        import sys
+
         _prepend_user_install_bins_to_path()
         importlib.invalidate_caches()
         try:
@@ -905,26 +991,115 @@ def setup_shell_routes() -> APIRouter:
                 raise HTTPException(400, "Invalid ssh_port")
         packages = [
             # ── System ── OS binaries, not pip packages
-            {"name": "tmux", "pip": "", "desc": "Required for Linux/Termux Cookbook background downloads and serves", "category": "System", "target": "remote", "kind": "system", "install_hint": "Run Cookbook server setup, or install tmux with apt/pacman/dnf/apk/zypper."},
-            {"name": "docker", "pip": "", "desc": "Required only for Docker-backed launch commands", "category": "System", "target": "remote", "kind": "system", "install_hint": "Install Docker on the selected server and allow this user to run docker."},
+            {
+                "name": "tmux",
+                "pip": "",
+                "desc": "Required for Linux/Termux Cookbook background downloads and serves",
+                "category": "System",
+                "target": "remote",
+                "kind": "system",
+                "install_hint": "Run Cookbook server setup, or install tmux with apt/pacman/dnf/apk/zypper.",
+            },
+            {
+                "name": "docker",
+                "pip": "",
+                "desc": "Required only for Docker-backed launch commands",
+                "category": "System",
+                "target": "remote",
+                "kind": "system",
+                "install_hint": "Install Docker on the selected server and allow this user to run docker.",
+            },
             # ── LLM ── installs on GPU servers for model serving/downloading
-            {"name": "hf_transfer", "pip": "hf_transfer", "desc": "Fast model downloads from HuggingFace", "category": "LLM", "target": "remote"},
-            {"name": "llama_cpp", "pip": "llama-cpp-python[server]", "desc": "Serve GGUF models via llama.cpp", "category": "LLM", "target": "remote"},
-            {"name": "sglang", "pip": "sglang[all]", "desc": "Serve HF safetensors models via SGLang", "category": "LLM", "target": "remote"},
-            {"name": "vllm", "pip": "vllm", "desc": "High-throughput LLM serving engine", "category": "LLM", "target": "remote"},
+            {
+                "name": "hf_transfer",
+                "pip": "hf_transfer",
+                "desc": "Fast model downloads from HuggingFace",
+                "category": "LLM",
+                "target": "remote",
+            },
+            {
+                "name": "llama_cpp",
+                "pip": "llama-cpp-python[server]",
+                "desc": "Serve GGUF models via llama.cpp",
+                "category": "LLM",
+                "target": "remote",
+            },
+            {
+                "name": "sglang",
+                "pip": "sglang[all]",
+                "desc": "Serve HF safetensors models via SGLang",
+                "category": "LLM",
+                "target": "remote",
+            },
+            {
+                "name": "vllm",
+                "pip": "vllm",
+                "desc": "High-throughput LLM serving engine",
+                "category": "LLM",
+                "target": "remote",
+            },
+            {
+                "name": "APFEL",
+                "pip": "",
+                "desc": "OpenAI-compatible API for Apple Foundational Models on Apple Silicon",
+                "category": "LLM",
+                "target": "local",
+                "kind": "system",
+                "install_cmd": "brew install apfel",
+                "update_cmd": "brew upgrade apfel",
+                "install_hint": "Requires a native Apple Silicon Mac with Apple Foundational Models support. Installable via Homebrew on supported Macs.",
+            },
             # ── Image ── editor + diffusion model serving
-            {"name": "diffusers", "pip": "diffusers[torch]", "desc": "Image generation pipelines (SD, Flux) with PyTorch", "category": "Image", "target": "remote"},
-            {"name": "rembg", "pip": "rembg[gpu]", "desc": "AI background removal for image editor", "category": "Image", "target": "local"},
-            {"name": "realesrgan", "pip": "realesrgan", "desc": "AI denoise + upscale (Real-ESRGAN). Used by editor's Denoise and Upscale tools.", "category": "Image", "target": "local"},
+            {
+                "name": "diffusers",
+                "pip": "diffusers[torch]",
+                "desc": "Image generation pipelines (SD, Flux) with PyTorch",
+                "category": "Image",
+                "target": "remote",
+            },
+            {
+                "name": "rembg",
+                "pip": "rembg[gpu]",
+                "desc": "AI background removal for image editor",
+                "category": "Image",
+                "target": "local",
+            },
+            {
+                "name": "realesrgan",
+                "pip": "realesrgan",
+                "desc": "AI denoise + upscale (Real-ESRGAN). Used by editor's Denoise and Upscale tools.",
+                "category": "Image",
+                "target": "local",
+            },
             # ── Tools ──
-            {"name": "playwright", "pip": "playwright", "desc": "Browser automation for web tools", "category": "Tools", "target": "local"},
+            {
+                "name": "playwright",
+                "pip": "playwright",
+                "desc": "Browser automation for web tools",
+                "category": "Tools",
+                "target": "local",
+            },
         ]
+
+        # Most packages should not be installed through external means. Hence, set the default of the
+        # install_cmd and update_cmd to None, which indicates that the recommended way to install/update is through the Cookbook # server setup or pip. Only system packages, should have explicit install/update commands provided.
+        for pkg in packages:
+            pkg.setdefault("install_cmd", None)
+            pkg.setdefault("update_cmd", None)
         # Remote check: for remote-target packages, probe the selected server's
         # venv over SSH so a remote `pip install` actually reflects here.
         remote_status: dict = {}
         remote_details: dict = {}
-        remote_names = [p["name"] for p in packages if p.get("target") == "remote" and p.get("kind") != "system"]
-        remote_system_names = [p["name"] for p in packages if p.get("target") == "remote" and p.get("kind") == "system"]
+        remote_names = [
+            p["name"]
+            for p in packages
+            if p.get("target") == "remote" and p.get("kind") != "system"
+        ]
+        remote_system_names = [
+            p["name"]
+            for p in packages
+            if p.get("target") == "remote" and p.get("kind") == "system"
+        ]
         if host and remote_names:
             try:
                 py = _package_probe_script(remote_names)
@@ -934,7 +1109,9 @@ def setup_shell_routes() -> APIRouter:
                 inner = f"{src}python3 -c {shlex.quote(py)}"
                 argv = _ssh_base_argv(host, ssh_port) + [inner]
                 proc = await asyncio.create_subprocess_exec(
-                    *argv, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE
+                    *argv,
+                    stdout=asyncio.subprocess.PIPE,
+                    stderr=asyncio.subprocess.PIPE,
                 )
                 out, _err = await asyncio.wait_for(proc.communicate(), timeout=12)
                 txt = out.decode("utf-8", errors="replace").strip()
@@ -958,11 +1135,15 @@ def setup_shell_routes() -> APIRouter:
                 checks = []
                 for name in remote_system_names:
                     qn = shlex.quote(name)
-                    checks.append(f"if command -v {qn} >/dev/null 2>&1; then echo {qn}=1; else echo {qn}=0; fi")
+                    checks.append(
+                        f"if command -v {qn} >/dev/null 2>&1; then echo {qn}=1; else echo {qn}=0; fi"
+                    )
                 inner = " ; ".join(checks)
                 argv = _ssh_base_argv(host, ssh_port) + [inner]
                 proc = await asyncio.create_subprocess_exec(
-                    *argv, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE
+                    *argv,
+                    stdout=asyncio.subprocess.PIPE,
+                    stderr=asyncio.subprocess.PIPE,
                 )
                 out, _err = await asyncio.wait_for(proc.communicate(), timeout=12)
                 txt = out.decode("utf-8", errors="replace").strip()
@@ -987,11 +1168,25 @@ def setup_shell_routes() -> APIRouter:
                     if note:
                         pkg["status_note"] = note
             elif pkg.get("kind") == "system":
-                pkg["installed"] = shutil.which(pkg["name"]) is not None
+                if pkg["name"] == "APFEL":
+                    pkg["applicable"] = IS_APPLE_SILICON
+                    pkg["installed"] = which_tool("apfel") is not None
+                    pkg["status_note"] = (
+                        "Available on Apple Silicon (arm64) devices; exposed through a local OpenAI-compatible API."
+                        if IS_APPLE_SILICON
+                        else "Requires a native Apple Silicon Mac with Apple Foundational Models support."
+                    )
+                else:
+                    pkg["installed"] = shutil.which(pkg["name"]) is not None
             elif pkg["name"] == "llama_cpp" and shutil.which("llama-server"):
                 pkg["installed"] = True
-                pkg["status_note"] = f"native llama-server: {shutil.which('llama-server')}"
-                probe = {"binaries": {"llama-server": shutil.which("llama-server")}, "dists": {}}
+                pkg["status_note"] = (
+                    f"native llama-server: {shutil.which('llama-server')}"
+                )
+                probe = {
+                    "binaries": {"llama-server": shutil.which("llama-server")},
+                    "dists": {},
+                }
             elif pkg["name"] == "vllm":
                 _vllm_cli = shutil.which("vllm")
                 pkg["installed"] = _vllm_cli is not None
@@ -1014,6 +1209,12 @@ def setup_shell_routes() -> APIRouter:
                     pkg["installed"] = False
                 except importlib_metadata.PackageNotFoundError:
                     pkg["installed"] = False
+                except Exception:
+                    # Installed but crashes on import — e.g. a CUDA build of
+                    # llama-cpp-python raising FileNotFoundError when the CUDA
+                    # toolkit dir is absent. One broken optional package must not
+                    # 500 the entire packages panel; report it as not usable.
+                    pkg["installed"] = False
 
             if pkg.get("installed"):
                 update_status = _package_pip_update_status(pkg, probe)
@@ -1037,15 +1238,30 @@ def setup_shell_routes() -> APIRouter:
         """Install a package via pip. Admin only — pip install is effectively code exec."""
         _require_admin(request)
         import sys as _sys
+
         body = await request.json()
         pip_name = body.get("pip")
         if not pip_name:
             return {"ok": False, "error": "No package specified"}
         # Validate against known packages to prevent arbitrary pip install
         known = {
-            "rembg[gpu]", "hf_transfer", "llama-cpp-python[server]", "sglang[all]", "diffusers", "diffusers[torch]",
-            "TTS", "bark", "faster-whisper", "playwright", "realesrgan", "gfpgan",
-            "insightface", "onnxruntime-gpu", "onnxruntime", "hdbscan", "vllm",
+            "rembg[gpu]",
+            "hf_transfer",
+            "llama-cpp-python[server]",
+            "sglang[all]",
+            "diffusers",
+            "diffusers[torch]",
+            "TTS",
+            "bark",
+            "faster-whisper",
+            "playwright",
+            "realesrgan",
+            "gfpgan",
+            "insightface",
+            "onnxruntime-gpu",
+            "onnxruntime",
+            "hdbscan",
+            "vllm",
         }
         if pip_name not in known:
             return {"ok": False, "error": f"Unknown package: {pip_name}"}
@@ -1071,6 +1287,7 @@ def setup_shell_routes() -> APIRouter:
         """
         _require_admin(request)
         from routes.cookbook_helpers import _llama_cpp_rebuild_cmd
+
         body = await request.json()
         engine = str(body.get("engine") or "llamacpp").strip()
         if engine != "llamacpp":
@@ -1079,7 +1296,11 @@ def setup_shell_routes() -> APIRouter:
         ssh_port = body.get("ssh_port")
         cmd = _llama_cpp_rebuild_cmd()
         try:
-            argv = (_ssh_base_argv(host, ssh_port) + [cmd]) if host else ["bash", "-lc", cmd]
+            argv = (
+                (_ssh_base_argv(host, ssh_port) + [cmd])
+                if host
+                else ["bash", "-lc", cmd]
+            )
         except ValueError as e:
             raise HTTPException(400, str(e))
         try:

routes/signature_routes.py

@@ -21,10 +21,44 @@ from src.auth_helpers import get_current_user
 logger = logging.getLogger(__name__)
 
 
-_DATA_URL_RE = re.compile(
-    r'^data:image/(?P<fmt>png|jpeg|jpg);base64,(?P<data>.+)$',
-    re.IGNORECASE | re.DOTALL,
-)
+_DATA_URL_RE = re.compile(r"^data:image/png;base64,(?P<data>.+)$", re.IGNORECASE | re.DOTALL)
+_ANY_IMAGE_DATA_URL_RE = re.compile(r"^data:image/[^;]+;base64,", re.IGNORECASE)
+_PNG_MAGIC = b"\x89PNG\r\n\x1a\n"
+_MAX_SIGNATURE_BYTES = 2 * 1024 * 1024
+_MAX_SIGNATURE_B64 = ((_MAX_SIGNATURE_BYTES + 2) // 3) * 4
+_MAX_SIGNATURE_DIMENSION = 4096
+
+
+def _normalize_signature_png(raw: str) -> str:
+    raw = (raw or "").strip()
+    m = _DATA_URL_RE.match(raw)
+    if m:
+        b64 = m.group("data")
+    elif _ANY_IMAGE_DATA_URL_RE.match(raw):
+        raise HTTPException(400, "Signature data must be a PNG image")
+    else:
+        b64 = raw
+    if len(b64) > _MAX_SIGNATURE_B64:
+        raise HTTPException(400, "Signature PNG is too large")
+    try:
+        payload = base64.b64decode(b64, validate=True)
+    except Exception:
+        raise HTTPException(400, "Signature data must be base64-encoded PNG bytes")
+    if not payload:
+        raise HTTPException(400, "Signature PNG is empty")
+    if len(payload) > _MAX_SIGNATURE_BYTES:
+        raise HTTPException(400, "Signature PNG is too large")
+    if not payload.startswith(_PNG_MAGIC):
+        raise HTTPException(400, "Signature data must be a PNG image")
+    return base64.b64encode(payload).decode("ascii")
+
+
+def _signature_dimension(value: Optional[int]) -> Optional[int]:
+    if value is None:
+        return None
+    if not isinstance(value, int) or value < 1 or value > _MAX_SIGNATURE_DIMENSION:
+        raise HTTPException(400, "Signature dimensions are invalid")
+    return value
 
 
 class SignatureCreate(BaseModel):
@@ -67,24 +101,18 @@ def setup_signature_routes() -> APIRouter:
     @router.post("/api/signatures")
     async def create_signature(request: Request, req: SignatureCreate) -> Dict[str, Any]:
         user = get_current_user(request)
-        raw = (req.data or "").strip()
-        m = _DATA_URL_RE.match(raw)
-        b64 = m.group("data") if m else raw
-        try:
-            payload = base64.b64decode(b64, validate=True)
-            if not payload:
-                raise ValueError("empty payload")
-        except Exception:
-            raise HTTPException(400, "Signature data must be base64-encoded PNG bytes")
+        b64 = _normalize_signature_png(req.data)
+        width = _signature_dimension(req.width)
+        height = _signature_dimension(req.height)
 
         sig = Signature(
             id=str(uuid.uuid4()),
             owner=user,
             name=(req.name or "Signature").strip() or "Signature",
             data_png=b64,
-            width=req.width,
-            height=req.height,
-            svg=req.svg,
+            width=width,
+            height=height,
+            svg=None,
         )
         db = SessionLocal()
         try:

routes/skills_routes.py

@@ -11,6 +11,8 @@ import logging
 import re
 from typing import List, Optional
 
+import httpx
+
 from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel, Field
 
@@ -51,6 +53,10 @@ class SkillAddRequest(BaseModel):
     steps: List[str] = Field(default_factory=list)
 
 
+class SkillImportUrlRequest(BaseModel):
+    url: str = Field(..., min_length=8, max_length=2000)
+
+
 class SkillUpdateRequest(BaseModel):
     name: Optional[str] = None
     description: Optional[str] = None
@@ -1014,7 +1020,7 @@ def _resolve_audit_models(owner=None):
             spec = (get_setting("teacher_model", "") or "").strip()
             if spec:
                 from src.ai_interaction import _resolve_model
-                t_url, t_model, t_headers = _resolve_model(spec)
+                t_url, t_model, t_headers = _resolve_model(spec, owner=owner)
                 if t_url and t_model:
                     teacher = (t_url, t_model, t_headers)
     except Exception as e:
@@ -1103,6 +1109,35 @@ def setup_skills_routes(skills_manager: SkillsManager) -> APIRouter:
         idx = skills_manager.index_for(owner=user)
         return {"index": idx, "count": len(idx)}
 
+    @router.get("/slash-catalog")
+    async def get_slash_catalog(request: Request):
+        """Return skills that are available as slash commands.
+
+        Mirrors the agent prompt's published-skill index so the UI never offers
+        a slash command the model would not normally be allowed to discover.
+        """
+        user = _owner(request)
+        all_skills = {s.get("name"): s for s in skills_manager.load(owner=user)}
+        entries = []
+        for s in skills_manager.index_for(owner=user):
+            name = (s.get("name") or "").strip()
+            if not name:
+                continue
+            full = all_skills.get(name) or {}
+            category = (s.get("category") or full.get("category") or "general").strip() or "general"
+            entries.append({
+                "type": "skill",
+                "token": f"/{name}",
+                "name": name,
+                "category": f"Skills / {category}",
+                "help": s.get("description") or full.get("description") or "",
+                "usage": f"/{name} <request>",
+                "uses": int(full.get("uses") or 0),
+                "last_used": full.get("last_used"),
+            })
+        entries.sort(key=lambda row: row["name"])
+        return {"skills": entries, "count": len(entries)}
+
     @router.get("/builtin")
     async def list_builtin_skills(request: Request):
         """Read-only list of the agent's built-in tool capabilities (research,
@@ -1203,6 +1238,36 @@ def setup_skills_routes(skills_manager: SkillsManager) -> APIRouter:
             save_settings(settings)
         return {"ok": True, "name": name, "is_overridden": False}
 
+    @router.post("/import-from-url")
+    async def import_skill_from_url(request: Request, body: SkillImportUrlRequest):
+        """Install a SKILL.md bundle from a public GitHub URL (skills.sh links supported)."""
+        require_admin(request)
+        user = _owner(request)
+        from services.memory.skill_importer import (
+            SkillImportError,
+            fetch_skill_bundle,
+        )
+
+        try:
+            files, _src = fetch_skill_bundle(body.url.strip())
+            entry = skills_manager.import_bundle_from_files(
+                files,
+                owner=user,
+                source_url=body.url.strip(),
+            )
+        except SkillImportError as e:
+            raise HTTPException(400, str(e)) from e
+        except httpx.HTTPError as e:
+            logger.warning("skill import fetch failed: %s", e)
+            detail = str(e).strip() or "Could not download skill from URL"
+            raise HTTPException(502, detail) from e
+        except Exception as e:
+            logger.error("skill import failed: %s", e)
+            raise HTTPException(500, "Skill import failed") from e
+
+        _fire_skill_added(user)
+        return {"ok": True, "skill": entry, "files": len(files)}
+
     @router.post("/add")
     async def add_skill(request: Request, body: SkillAddRequest):
         user = _owner(request)
@@ -1236,6 +1301,47 @@ def setup_skills_routes(skills_manager: SkillsManager) -> APIRouter:
             _fire_skill_added(user)
         return {"ok": True, "deduped": bool(entry.get("_deduped")), "skill": entry}
 
+    @router.post("/{skill_id}/invoke")
+    async def invoke_skill(request: Request, skill_id: str):
+        """Build a skill-pinned prompt for slash-command invocation.
+
+        This is intentionally server-side so availability, ownership, and usage
+        accounting use the same rules as the SkillsManager.
+        """
+        user = _owner(request)
+        try:
+            body = await request.json()
+        except Exception:
+            body = {}
+        request_text = (body.get("request") or "").strip() if isinstance(body, dict) else ""
+
+        invokable = {
+            s.get("name"): s for s in skills_manager.index_for(owner=user)
+            if (s.get("name") or "").strip()
+        }
+        match = invokable.get(skill_id)
+        if not match:
+            raise HTTPException(404, "Skill is not available for slash invocation")
+
+        name = match.get("name")
+        md = skills_manager.read_skill_md(name, owner=user)
+        if md is None:
+            raise HTTPException(404, "Skill source unavailable")
+
+        skills_manager.record_use(name, owner=user)
+        message = (
+            "Apply the skill below to my request, following its Procedure / Pitfalls / Verification.\n\n"
+            f"--- BEGIN SKILL ---\n{md}\n--- END SKILL ---\n\n"
+            + (f"Request: {request_text}" if request_text else "Request: (use the skill as appropriate)")
+        )
+        return {
+            "ok": True,
+            "type": "skill",
+            "name": name,
+            "command": f"/{name}",
+            "message": message,
+        }
+
     @router.get("/{skill_id}")
     async def get_skill(request: Request, skill_id: str):
         user = _owner(request)

routes/stt_routes.py

@@ -4,12 +4,10 @@
 from fastapi import APIRouter, HTTPException, UploadFile, File
 import logging
 
-from src.upload_limits import read_upload_limited
+from src.upload_limits import read_upload_limited, STT_MAX_AUDIO_BYTES
 
 logger = logging.getLogger(__name__)
 
-STT_MAX_AUDIO_BYTES = 25 * 1024 * 1024
-
 
 def setup_stt_routes(stt_service):
     """Setup STT routes with the provided STT service"""

routes/task_routes.py

@@ -11,13 +11,128 @@ from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel
 
 from core.database import SessionLocal, ScheduledTask, TaskRun
+from core.constants import internal_api_base
 from src.auth_helpers import get_current_user
+from src.constants import DATA_DIR, EMAIL_URGENCY_CACHE_DIR
 from src.task_scheduler import compute_next_run, HOUSEKEEPING_DEFAULTS
 from routes.prefs_routes import _load_for_user, _save_for_user
 
 logger = logging.getLogger(__name__)
 
 
+def _maybe_cascade_calendar_event(task) -> None:
+    """Delete the linked calendar event when a cookbook_serve task is
+    removed. Two lookup strategies:
+
+      1. PRIMARY — `cookbook_event_uid` marker stashed in task.prompt
+         by cookbookSchedule.js right after creating the event. Direct
+         UID match, no ambiguity.
+
+      2. FALLBACK — for tasks created before the marker was wired up
+         (or when the PATCH to add the marker failed silently), scan
+         the Cookbook calendar for events whose summary equals the
+         task name and delete the matches.
+
+    Best-effort throughout: errors are logged but never block the task
+    deletion itself."""
+    if not task or task.task_type != "action" or task.action != "cookbook_serve":
+        return
+
+    import httpx
+    from core.middleware import INTERNAL_TOOL_HEADER, INTERNAL_TOOL_TOKEN
+    headers = {INTERNAL_TOOL_HEADER: INTERNAL_TOOL_TOKEN}
+    if task.owner:
+        headers["X-Odysseus-Owner"] = task.owner
+
+    # Strategy 1: explicit UID marker in prompt.
+    event_uid = ""
+    if task.prompt:
+        try:
+            cfg = json.loads(task.prompt)
+            if isinstance(cfg, dict):
+                event_uid = (cfg.get("cookbook_event_uid") or "").strip()
+        except Exception:
+            pass
+
+    def _try_delete(uid: str) -> bool:
+        try:
+            with httpx.Client(timeout=10) as client:
+                r = client.delete(
+                    f"{internal_api_base()}/api/calendar/events/{uid}",
+                    headers=headers,
+                )
+                if r.status_code >= 400:
+                    logger.info(
+                        f"task delete: cascade calendar event {uid} returned "
+                        f"HTTP {r.status_code}"
+                    )
+                    return False
+                return True
+        except Exception as e:
+            logger.warning(f"task delete: cascade calendar event {uid} failed: {e}")
+            return False
+
+    if event_uid:
+        _try_delete(event_uid)
+        return
+
+    # Strategy 2: scan the Cookbook calendar for matching summaries.
+    # Only runs for tasks missing the marker (old tasks or PATCH failures).
+    if not task.name:
+        return
+    try:
+        with httpx.Client(timeout=10) as client:
+            # Find the Cookbook calendar.
+            cal_r = client.get(f"{internal_api_base()}/api/calendar/calendars", headers=headers)
+            if cal_r.status_code >= 400:
+                return
+            cals = (cal_r.json() or {}).get("calendars", [])
+            cookbook_cal = next(
+                (c for c in cals if (c.get("name") or "").lower() == "cookbook"),
+                None,
+            )
+            if not cookbook_cal:
+                return
+            cal_href = cookbook_cal.get("href") or cookbook_cal.get("id") or ""
+            # List events in a wide window to catch recurring + upcoming.
+            from datetime import datetime as _dt, timedelta as _td, timezone as _tz
+            now = _dt.now(_tz.utc)
+            start = (now - _td(days=30)).isoformat()
+            end = (now + _td(days=365)).isoformat()
+            ev_r = client.get(
+                f"{internal_api_base()}/api/calendar/events",
+                params={"start": start, "end": end, "calendar": cal_href},
+                headers=headers,
+            )
+            if ev_r.status_code >= 400:
+                return
+            events = (ev_r.json() or {}).get("events", [])
+            # Match by exact summary. Tasks named "Serve: <model>" are
+            # created from the schedule modal; the event's summary mirrors
+            # the task name 1:1 by design.
+            target = (task.name or "").strip()
+            uids_to_delete = set()
+            for ev in events:
+                if (ev.get("summary") or "").strip() != target:
+                    continue
+                uid = ev.get("uid") or ev.get("id") or ""
+                # Strip the "::occurrence" suffix on recurring expansions —
+                # we want to delete the MASTER once, not each instance.
+                if "::" in uid:
+                    uid = uid.split("::", 1)[0]
+                if uid:
+                    uids_to_delete.add(uid)
+            for uid in uids_to_delete:
+                _try_delete(uid)
+            if uids_to_delete:
+                logger.info(
+                    f"task delete: cascade matched {len(uids_to_delete)} calendar event(s) "
+                    f"by summary fallback for task {task.id} ({target!r})"
+                )
+    except Exception as e:
+        logger.warning(f"task delete: cascade fallback scan failed: {e}")
+
+
 class TaskCreate(BaseModel):
     name: Optional[str] = None
     prompt: Optional[str] = None
@@ -178,20 +293,24 @@ def setup_task_routes(task_scheduler) -> APIRouter:
     def _owner(request: Request):
         return get_current_user(request)
 
-    async def _generate_task_name(prompt: str) -> str:
+    async def _generate_task_name(prompt: str, owner: Optional[str] = None) -> str:
         """Use LLM to generate a short task name from the prompt."""
         try:
             from src.llm_core import llm_call_async
             from core.database import Session as DbSession
             db = SessionLocal()
             try:
-                recent = db.query(DbSession).filter(
+                q = db.query(DbSession).filter(
                     DbSession.endpoint_url.isnot(None),
                     DbSession.model.isnot(None),
-                ).order_by(DbSession.created_at.desc()).first()
+                )
+                if owner:
+                    q = q.filter(DbSession.owner == owner)
+                recent = q.order_by(DbSession.created_at.desc()).first()
                 if not recent:
                     return prompt[:50].strip()
                 url, model = recent.endpoint_url, recent.model
+                headers = recent.headers or {}
             finally:
                 db.close()
 
@@ -202,6 +321,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                     {"role": "user", "content": prompt[:500]},
                 ],
                 max_tokens=20,
+                headers=headers,
                 timeout=15,
             )
             title = result.strip().strip('"\'').strip()
@@ -316,6 +436,20 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         except Exception:
             return False
 
+    def _validate_then_task_id(db, then_task_id: Optional[str], user: Optional[str], current_task_id: Optional[str] = None) -> Optional[str]:
+        target_id = (then_task_id or "").strip()
+        if not target_id:
+            return None
+        if current_task_id and target_id == current_task_id:
+            raise HTTPException(400, "Task cannot chain to itself")
+        q = db.query(ScheduledTask).filter(ScheduledTask.id == target_id)
+        if user:
+            q = q.filter(ScheduledTask.owner == user)
+        target = q.first()
+        if not target:
+            raise HTTPException(404, "Chained task not found")
+        return target.id
+
     @router.post("")
     async def create_task(request: Request, req: TaskCreate):
         user = _owner(request)
@@ -352,7 +486,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 from src.builtin_actions import BUILTIN_ACTION_INFO
                 name = BUILTIN_ACTION_INFO.get(req.action, req.action or "Action Task")
             elif req.prompt:
-                name = await _generate_task_name(req.prompt)
+                name = await _generate_task_name(req.prompt, owner=user)
             else:
                 name = "Untitled Task"
 
@@ -379,11 +513,21 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         task_id = str(uuid.uuid4())
         db = SessionLocal()
         try:
+            then_task_id = _validate_then_task_id(db, req.then_task_id, user)
             notifications_enabled = (
                 False if req.task_type == "action" and req.notifications_enabled is None
                 else bool(req.notifications_enabled) if req.notifications_enabled is not None
                 else True
             )
+            # Validate chained task belongs to same owner
+            if req.then_task_id:
+                chain_target = db.query(ScheduledTask).filter(
+                    ScheduledTask.id == req.then_task_id
+                ).first()
+                if not chain_target:
+                    raise HTTPException(400, "Chained task not found")
+                if chain_target.owner != user:
+                    raise HTTPException(403, "Cannot chain to another user's task")
             task = ScheduledTask(
                 id=task_id,
                 owner=user,
@@ -405,7 +549,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 output_target=req.output_target,
                 model=req.model or None,
                 endpoint_url=req.endpoint_url or None,
-                then_task_id=req.then_task_id or None,
+                then_task_id=then_task_id,
                 webhook_token=webhook_token,
                 notifications_enabled=notifications_enabled,
             )
@@ -487,7 +631,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
 
         removed_files = 0
         if action == "check_email_urgency":
-            cache_dir = Path("data/email_urgency_cache")
+            cache_dir = Path(EMAIL_URGENCY_CACHE_DIR)
             if cache_dir.exists():
                 for child in cache_dir.glob("*.json"):
                     try:
@@ -496,7 +640,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                     except Exception:
                         pass
             owner_slug = "".join(c if (c.isalnum() or c in "-_.@") else "_" for c in (user or "default"))
-            for state_path in [Path(f"data/email_urgency_state_{owner_slug}.json")]:
+            for state_path in [Path(DATA_DIR) / f"email_urgency_state_{owner_slug}.json"]:
                 try:
                     if state_path.exists():
                         state_path.unlink()
@@ -558,7 +702,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             if req.trigger_count is not None:
                 task.trigger_count = req.trigger_count
             if req.then_task_id is not None:
-                task.then_task_id = req.then_task_id or None
+                task.then_task_id = _validate_then_task_id(db, req.then_task_id, user, current_task_id=task.id)
             if req.notifications_enabled is not None:
                 task.notifications_enabled = bool(req.notifications_enabled)
             if req.cron_expression is not None:
@@ -616,6 +760,12 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 raise HTTPException(404, "Task not found")
             if user and task.owner != user:
                 raise HTTPException(403, "Access denied")
+            # Cascade: cookbook_serve tasks may have a linked calendar
+            # event (created via the "Create event in calendar" toggle
+            # in the schedule modal). If so, delete the calendar event
+            # too so the calendar doesn't end up holding a phantom event
+            # for a task that no longer exists.
+            _maybe_cascade_calendar_event(task)
             db.delete(task)
             db.commit()
             return {"ok": True}
@@ -833,7 +983,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             "tag", "label", "move", "archive", "delete", "mark", "schedule",
         )
         try:
-            from src.agent_tools import get_mcp_manager
+            from src.tool_utils import get_mcp_manager
             mcp = get_mcp_manager()
             if mcp:
                 for tool in mcp.get_all_tools():
@@ -928,6 +1078,7 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         desc = (body.get("description") or "").strip()
         if not desc:
             return {"success": False, "message": "Nothing to parse"}
+        user = _owner(request)
 
         now = _dt.now()
         # Give the model the current date/time + weekday so relative phrasing
@@ -954,9 +1105,9 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             "use cron '0 H * * 1-5'. Keep the prompt actionable and self-contained."
         )
         try:
-            url, model, headers = resolve_endpoint("utility")
+            url, model, headers = resolve_endpoint("utility", owner=user or None)
             if not url:
-                url, model, headers = resolve_endpoint("default")
+                url, model, headers = resolve_endpoint("default", owner=user or None)
             if not (url and model):
                 return {"success": False, "message": "No model endpoint configured"}
             raw = await llm_call_async(

routes/upload_routes.py

@@ -13,9 +13,43 @@ from src.upload_handler import count_recent_uploads
 logger = logging.getLogger(__name__)
 
 router = APIRouter(prefix="/api/upload", tags=["upload"])
+UPLOAD_RESPONSE_HEADERS = {"X-Content-Type-Options": "nosniff"}
 
 def setup_upload_routes(upload_handler):
     """Setup upload routes with the provided handler"""
+
+    def _upload_root() -> str:
+        from src.constants import UPLOAD_DIR
+        return os.path.realpath(getattr(upload_handler, "upload_dir", UPLOAD_DIR))
+
+    def _path_inside_upload_dir(path: str) -> bool:
+        try:
+            return os.path.commonpath([_upload_root(), os.path.realpath(path)]) == _upload_root()
+        except Exception:
+            return False
+
+    def _resolve_upload_path(file_id: str) -> str:
+        from src.constants import UPLOAD_DIR
+        upload_root = getattr(upload_handler, "upload_dir", UPLOAD_DIR)
+        direct = os.path.join(upload_root, file_id)
+        if os.path.lexists(direct):
+            if not _path_inside_upload_dir(direct):
+                raise HTTPException(403, "Access denied")
+            if os.path.isfile(direct):
+                return direct
+            raise HTTPException(404, "File not found")
+
+        for root, _dirs, files in os.walk(upload_root, followlinks=False):
+            if file_id not in files:
+                continue
+            path = os.path.join(root, file_id)
+            if not _path_inside_upload_dir(path):
+                raise HTTPException(403, "Access denied")
+            if os.path.isfile(path):
+                return path
+            raise HTTPException(404, "File not found")
+
+        raise HTTPException(404, "File not found")
     
     @router.post("")
     async def api_upload(request: Request, files: List[UploadFile] = File(...)):
@@ -91,23 +125,11 @@ def setup_upload_routes(upload_handler):
         client isn't downloading the full-resolution photo just to show it tiny."""
         if not upload_handler.validate_upload_id(file_id):
             raise HTTPException(400, "Invalid file ID")
-        # Search upload directories for the file
-        from src.constants import UPLOAD_DIR
         import mimetypes as _mt
-        path = os.path.join(UPLOAD_DIR, file_id)
-        if not os.path.exists(path):
-            for root, dirs, files in os.walk(UPLOAD_DIR):
-                if file_id in files:
-                    path = os.path.join(root, file_id)
-                    break
-            else:
-                raise HTTPException(404, "File not found")
-        if not upload_handler.inside_base_dir(path):
-            raise HTTPException(403, "Access denied")
         # Look up original filename and owner from uploads.json
         original_name = file_id
         info = None
-        uploads_db = os.path.join(UPLOAD_DIR, "uploads.json")
+        uploads_db = os.path.join(_upload_root(), "uploads.json")
         if os.path.exists(uploads_db):
             with open(uploads_db, encoding="utf-8") as f:
                 db = json.load(f)
@@ -123,13 +145,14 @@ def setup_upload_routes(upload_handler):
                 raise HTTPException(403, "Access denied")
             if file_owner != current_user and not auth_mgr.is_admin(current_user):
                 raise HTTPException(404, "File not found")
-        mime = _mt.guess_type(path)[0] or "application/octet-stream"
+        path = _resolve_upload_path(file_id)
+        mime = (info or {}).get("mime") or _mt.guess_type(path)[0] or "application/octet-stream"
         from fastapi.responses import FileResponse
         # Downscaled thumbnail for image previews — generated once and cached.
         if thumb and mime.startswith("image/"):
             try:
                 from PIL import Image, ImageOps
-                thumb_dir = os.path.join(UPLOAD_DIR, ".thumbs")
+                thumb_dir = os.path.join(_upload_root(), ".thumbs")
                 os.makedirs(thumb_dir, exist_ok=True)
                 thumb_path = os.path.join(thumb_dir, file_id + ".jpg")
                 if (not os.path.exists(thumb_path)
@@ -145,17 +168,21 @@ def setup_upload_routes(upload_handler):
                     if im.mode not in ("RGB", "L"):
                         im = im.convert("RGB")
                     im.save(thumb_path, "JPEG", quality=80)
-                return FileResponse(thumb_path, media_type="image/jpeg")
+                return FileResponse(thumb_path, media_type="image/jpeg", headers=UPLOAD_RESPONSE_HEADERS)
             except Exception as e:
                 logger.warning(f"Thumbnail generation failed for {file_id}: {e}")
                 # Fall through to the full image.
-        return FileResponse(path, media_type=mime, filename=original_name)
+        return FileResponse(
+            path,
+            media_type=mime,
+            filename=original_name,
+            headers=UPLOAD_RESPONSE_HEADERS,
+        )
 
     def _load_upload_info(file_id: str):
         """Look up the uploads.json record for a file_id, with owner/auth checks."""
-        from src.constants import UPLOAD_DIR
         info = None
-        uploads_db = os.path.join(UPLOAD_DIR, "uploads.json")
+        uploads_db = os.path.join(_upload_root(), "uploads.json")
         if os.path.exists(uploads_db):
             with open(uploads_db, encoding="utf-8") as f:
                 db = json.load(f)
@@ -163,8 +190,7 @@ def setup_upload_routes(upload_handler):
         return info
 
     def _vision_cache_path(file_id: str) -> str:
-        from src.constants import UPLOAD_DIR
-        cache_dir = os.path.join(UPLOAD_DIR, ".vision")
+        cache_dir = os.path.join(_upload_root(), ".vision")
         os.makedirs(cache_dir, exist_ok=True)
         return os.path.join(cache_dir, file_id + ".txt")
 
@@ -175,17 +201,6 @@ def setup_upload_routes(upload_handler):
         subsequent loads are instant. Pass force=1 to recompute."""
         if not upload_handler.validate_upload_id(file_id):
             raise HTTPException(400, "Invalid file ID")
-        from src.constants import UPLOAD_DIR
-        path = os.path.join(UPLOAD_DIR, file_id)
-        if not os.path.exists(path):
-            for root, dirs, files in os.walk(UPLOAD_DIR):
-                if file_id in files:
-                    path = os.path.join(root, file_id)
-                    break
-            else:
-                raise HTTPException(404, "File not found")
-        if not upload_handler.inside_base_dir(path):
-            raise HTTPException(403, "Access denied")
         info = _load_upload_info(file_id)
         auth_mgr = getattr(request.app.state, "auth_manager", None)
         auth_configured = bool(auth_mgr and auth_mgr.is_configured)
@@ -196,8 +211,9 @@ def setup_upload_routes(upload_handler):
                 raise HTTPException(403, "Access denied")
             if file_owner != current_user and not auth_mgr.is_admin(current_user):
                 raise HTTPException(404, "File not found")
+        path = _resolve_upload_path(file_id)
         import mimetypes as _mt
-        mime = _mt.guess_type(path)[0] or ""
+        mime = (info or {}).get("mime") or _mt.guess_type(path)[0] or ""
         if not mime.startswith("image/"):
             raise HTTPException(400, "Not an image")
         cache_path = _vision_cache_path(file_id)
@@ -209,7 +225,7 @@ def setup_upload_routes(upload_handler):
                 logger.warning(f"Vision cache read failed for {file_id}: {e}")
         from src.document_processor import analyze_image_with_vl
         try:
-            text = analyze_image_with_vl(path) or ""
+            text = analyze_image_with_vl(path, owner=current_user) or ""
         except Exception as e:
             logger.error(f"Vision analysis failed for {file_id}: {e}")
             raise HTTPException(500, f"Vision analysis failed: {e}")
@@ -238,6 +254,7 @@ def setup_upload_routes(upload_handler):
                 raise HTTPException(403, "Access denied")
             if file_owner != current_user and not auth_mgr.is_admin(current_user):
                 raise HTTPException(404, "File not found")
+        _resolve_upload_path(file_id)
         body = await request.json()
         text = (body or {}).get("text", "")
         if not isinstance(text, str):

routes/vault_routes.py

@@ -17,10 +17,11 @@ from pydantic import BaseModel
 
 from core.middleware import require_admin
 from core.platform_compat import IS_WINDOWS, safe_chmod, which_tool
+from src.constants import VAULT_FILE as _VAULT_FILE
 
 logger = logging.getLogger(__name__)
 
-VAULT_FILE = Path("data/vault.json")
+VAULT_FILE = Path(_VAULT_FILE)
 
 
 def _find_bw() -> str:

routes/webhook_routes.py

@@ -194,6 +194,8 @@ def setup_webhook_routes(
         "together": "https://api.together.xyz/v1",
         "openrouter": "https://openrouter.ai/api/v1",
         "ollama": "https://ollama.com/api",
+        "opencode-zen": "https://opencode.ai/zen/v1",
+        "opencode-go": "https://opencode.ai/zen/go/v1",
         "fireworks": "https://api.fireworks.ai/inference/v1",
         "venice": "https://api.venice.ai/api/v1",
     }
@@ -323,22 +325,33 @@ def setup_webhook_routes(
             endpoint_url = build_chat_url(base_url)
             model = body.model or "auto"
             api_key = ep.api_key
+            if getattr(ep, "provider_auth_id", None):
+                try:
+                    from src.endpoint_resolver import resolve_endpoint_runtime
+                    base_url, api_key = resolve_endpoint_runtime(ep, owner=token_owner)
+                    endpoint_url = build_chat_url(base_url)
+                except Exception:
+                    raise HTTPException(500, "Could not resolve endpoint credentials")
 
             if model == "auto":
                 try:
                     async with httpx.AsyncClient(timeout=5) as client:
                         models_url = build_models_url(base_url)
                         hdrs = build_headers(api_key, base_url)
-                        resp = await client.get(models_url, headers=hdrs)
-                        resp.raise_for_status()
-                        data = resp.json()
-                        ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
-                        if not ids:
-                            ids = [
-                                m.get("name") or m.get("model")
-                                for m in (data.get("models") or [])
-                                if m.get("name") or m.get("model")
-                            ]
+                        if models_url:
+                            resp = await client.get(models_url, headers=hdrs)
+                            resp.raise_for_status()
+                            data = resp.json()
+                            ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
+                            if not ids:
+                                ids = [
+                                    m.get("name") or m.get("model")
+                                    for m in (data.get("models") or [])
+                                    if m.get("name") or m.get("model")
+                                ]
+                        else:
+                            import json as _json
+                            ids = _json.loads(ep.cached_models or "[]")
                         model = ids[0] if ids else "auto"
                 except Exception:
                     raise HTTPException(500, "Could not discover models from endpoint")

routes/workspace_routes.py

@@ -1,10 +1,15 @@
-"""Workspace API — browse server directories to pick a tool workspace folder."""
+"""Workspace API - browse server directories to pick a tool workspace folder."""
 import os
 from fastapi import APIRouter, Request, HTTPException, Query
 
 from src.auth_helpers import get_current_user
 from src.tool_security import owner_is_admin_or_single_user
 
+# Cap entries returned per directory (mirrors filesystem_tools._CODENAV_MAX_HITS).
+# A huge directory shouldn't dump thousands of rows into the picker; the user can
+# type/paste a path to jump straight in instead.
+_MAX_BROWSE_DIRS = 500
+
 
 def setup_workspace_routes():
     router = APIRouter(prefix="/api/workspace", tags=["workspace"])
@@ -34,7 +39,7 @@ def setup_workspace_routes():
             with os.scandir(target) as it:
                 for entry in it:
                     try:
-                        # Don't follow symlinks when classifying — a symlinked
+                        # Don't follow symlinks when classifying - a symlinked
                         # dir is skipped rather than letting the browser wander
                         # off via a link. Hidden entries are omitted.
                         if entry.is_dir(follow_symlinks=False) and not entry.name.startswith("."):
@@ -46,11 +51,35 @@ def setup_workspace_routes():
         except (PermissionError, OSError):
             dirs = []
 
+        dirs_sorted = sorted(dirs, key=lambda d: d["name"].lower())
+        truncated = len(dirs_sorted) > _MAX_BROWSE_DIRS
         parent = os.path.dirname(target)
+        from src.tool_execution import vet_workspace
         return {
             "path": target,
             "parent": parent if parent and parent != target else None,
-            "dirs": sorted(dirs, key=lambda d: d["name"].lower()),
+            "dirs": dirs_sorted[:_MAX_BROWSE_DIRS],
+            "truncated": truncated,
+            # Whether this directory may be bound as a workspace (filesystem
+            # roots and sensitive dirs may be browsed through but not chosen).
+            "selectable": vet_workspace(target) is not None,
         }
 
+    @router.get("/vet")
+    def vet(request: Request, path: str = Query(default="")):
+        """Validate a workspace path without binding it.
+
+        The UI calls this before persisting a manually typed path (/workspace
+        set) so a typo, file path, deleted folder, sensitive dir, or filesystem
+        root is rejected up front with the canonical path returned on success,
+        instead of being stored client-side and silently dropped at chat time.
+        Admin-gated like /browse: it confirms path existence on the host.
+        """
+        owner = get_current_user(request)
+        if not owner_is_admin_or_single_user(owner):
+            raise HTTPException(status_code=403, detail="Workspace selection is admin-only")
+        from src.tool_execution import vet_workspace
+        resolved = vet_workspace(path)
+        return {"ok": resolved is not None, "path": resolved}
+
     return router

scripts/claim_ownerless.py

@@ -13,6 +13,8 @@ import json
 
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
+from src.constants import MEMORY_FILE, SKILLS_FILE
+
 
 def claim_json_entries(entries, owner):
     count = 0
@@ -35,8 +37,8 @@ def main():
 
     # 1. Memories (JSON files)
     for label, path in [
-        ("memory.json", "data/memory.json"),
-        ("skills.json", "data/skills.json"),
+        ("memory.json", MEMORY_FILE),
+        ("skills.json", SKILLS_FILE),
     ]:
         if not os.path.exists(path):
             print(f"  {label}: not found, skipping")

scripts/diffusion_server.py

@@ -34,6 +34,7 @@ import torch
 import uvicorn
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
+from starlette.middleware.trustedhost import TrustedHostMiddleware
 from pydantic import BaseModel
 
 logging.basicConfig(level=logging.INFO)
@@ -52,7 +53,63 @@ async def lifespan(application):
 
 
 app = FastAPI(title="Diffusion Server", lifespan=lifespan)
-app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+# Conservative defaults — server is designed for server-to-server use from
+# the Odysseus backend. Wildcard CORS + the 127.0.0.1 default bind used to
+# leave the server reachable via DNS-rebinding from any browser tab on the
+# same host. The CLI flags below extend these allowlists for operators who
+# need browser access; the safe defaults handle the common case.
+_DEFAULT_ALLOWED_HOSTS = ["127.0.0.1", "localhost", "::1"]
+_DEFAULT_CORS_ORIGINS: list = []  # default-deny
+
+
+def _compute_allowed_hosts(bind_host: str, extras=None) -> list:
+    """Allowed Host header values: the bind address + loopback variants +
+    any operator-supplied --allowed-host values. Duplicates and empty
+    strings are dropped; order is stable for predictable middleware setup."""
+    seen = []
+    for h in (bind_host, *_DEFAULT_ALLOWED_HOSTS, *(extras or [])):
+        h = (h or "").strip()
+        if h and h not in seen:
+            seen.append(h)
+    return seen
+
+
+def _compute_cors_origins(extras=None) -> list:
+    """CORS allowlist: default-deny (empty), extended only by explicit
+    --allowed-origin values. Server-to-server callers don't set an Origin
+    header so they're unaffected; this only narrows browser access."""
+    seen = []
+    for o in (*_DEFAULT_CORS_ORIGINS, *(extras or [])):
+        o = (o or "").strip()
+        if o and o not in seen:
+            seen.append(o)
+    return seen
+
+
+def _configure_security_middleware(application, allowed_hosts, allowed_origins):
+    """Replace `application`'s user middleware stack with the diffusion server
+    security middleware: the TrustedHost allowlist and, when origins are
+    supplied, CORS. Used at module load and by the __main__ CLI path before
+    serving starts. Raises before mutating if the middleware stack has already
+    been built. Order is preserved: TrustedHost first, then CORS (added last ->
+    outermost)."""
+    if application.middleware_stack is not None:
+        raise RuntimeError("security middleware must be configured before the app starts serving")
+    application.user_middleware.clear()
+    application.add_middleware(TrustedHostMiddleware, allowed_hosts=list(allowed_hosts))
+    if allowed_origins:
+        application.add_middleware(
+            CORSMiddleware,
+            allow_origins=list(allowed_origins),
+            allow_methods=["GET", "POST", "OPTIONS"],
+            allow_headers=["Authorization", "Content-Type"],
+        )
+
+
+# Install defaults at module load so importing the app for tests / direct
+# uvicorn invocation still benefits from the Host-header allowlist.
+_configure_security_middleware(app, _DEFAULT_ALLOWED_HOSTS, _DEFAULT_CORS_ORIGINS)
 
 
 class ImageRequest(BaseModel):
@@ -1089,7 +1146,25 @@ if __name__ == "__main__":
     parser.add_argument("--attention-slicing", action="store_true", help="Enable attention slicing")
     parser.add_argument("--vae-slicing", action="store_true", help="Enable VAE slicing")
     parser.add_argument("--harmonize-gpu", type=int, default=None, help="GPU index for harmonize/img2img (default: same as main)")
+    parser.add_argument("--allowed-host", action="append", default=[],
+        help="Additional Host header value to accept (DNS-rebinding allowlist). "
+             "Can be repeated. Loopback values are always included.")
+    parser.add_argument("--allowed-origin", action="append", default=[],
+        help="Additional CORS origin to allow. Can be repeated. Defaults to "
+             "no cross-origin access — only pass this if you need a browser "
+             "on a specific origin to call the server.")
     _args = parser.parse_args()
 
+    # Replace the module-load middleware stack with the CLI-configured one so
+    # operator-supplied --allowed-host / --allowed-origin values take effect
+    # before the first request is served. user_middleware is consulted lazily
+    # when the middleware stack is built on the first request, so mutating it
+    # here is safe.
+    final_hosts = _compute_allowed_hosts(_args.host, _args.allowed_host)
+    final_origins = _compute_cors_origins(_args.allowed_origin)
+    _configure_security_middleware(app, final_hosts, final_origins)
+    logger.info("security middleware: allowed_hosts=%s allowed_origins=%s",
+                final_hosts, final_origins or "(none — default-deny)")
+
     app.state.model_path = _args.model
     uvicorn.run(app, host=_args.host, port=_args.port)

scripts/index_documents.py

@@ -19,6 +19,9 @@ import sys
 from pathlib import Path
 from typing import List, Tuple
 
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from src.constants import PERSONAL_DIR
+
 # Configure logging for the script
 logging.basicConfig(
     level=logging.INFO,
@@ -45,7 +48,7 @@ def main():
     rag_manager = RAGManager()
     
     # Directory to scan
-    docs_directory = "data/personal_docs"
+    docs_directory = PERSONAL_DIR
     directory_path = Path(docs_directory)
     
     # Check if directory exists

scripts/migrate_faiss_to_chroma.py

@@ -63,10 +63,10 @@ def migrate_memories():
     """Migrate memory vectors from FAISS to ChromaDB."""
     from src.chroma_client import get_chroma_client
     from src.embeddings import get_embedding_client
-    from src.constants import DATA_DIR
+    from src.constants import MEMORY_VECTORS_DIR, MEMORY_FILE
 
-    ids_path = os.path.join(DATA_DIR, "memory_vectors", "ids.json")
-    memory_path = os.path.join(DATA_DIR, "memory.json")
+    ids_path = os.path.join(MEMORY_VECTORS_DIR, "ids.json")
+    memory_path = MEMORY_FILE
 
     if not os.path.exists(ids_path):
         logger.info("No memory FAISS index found, skipping memory migration")

scripts/odysseus-cookbook

@@ -47,6 +47,9 @@ _STATE_PATH = _DATA_DIR / "cookbook_state.json"
 import tempfile
 _TMUX_LOG_DIR = Path(tempfile.gettempdir()) / "odysseus-tmux"
 
+from core.platform_compat import NVIDIA_PATH_CANDIDATES, SSH_PATH_OVERRIDE
+
+
 
 def fail(msg: str, code: int = 1) -> None:
     sys.stderr.write(f"error: {msg}\n")
@@ -160,7 +163,26 @@ def cmd_gpus(args) -> None:
     prefix = _ssh_prefix(args.host, args.ssh_port)
     cmd = prefix + (query.split() if not prefix else [query])
     try:
-        out = subprocess.run(cmd, capture_output=True, text=True, timeout=15)
+        if prefix:
+            candidates = [query]
+            args_part = query[len("nvidia-smi "):]
+            candidates.append(
+                "bash -lc "
+                + repr(
+                    f"{SSH_PATH_OVERRIDE}"
+                    f"nvidia-smi {args_part}"
+                )
+            )
+            for nvidia_path in NVIDIA_PATH_CANDIDATES:
+                candidates.append(f"{nvidia_path} {args_part}")
+
+            out = None
+            for candidate in candidates:
+                out = subprocess.run(prefix + [candidate], capture_output=True, text=True, timeout=15)
+                if out.returncode == 0:
+                    break
+        else:
+            out = subprocess.run(cmd, capture_output=True, text=True, timeout=15)
     except FileNotFoundError:
         # No nvidia-smi locally → try the Metal fallback before giving up.
         if not prefix:

scripts/odysseus-research

@@ -25,6 +25,24 @@ from pathlib import Path
 
 _DATA_DIR = _REPO_ROOT / "data" / "deep_research"
 
+# The CLI's --status takes the user-facing label "complete", but the writer
+# in services/research/research_handler.py stores `status="done"` when a run
+# finishes (and the legacy src/research_handler.py does the same). Without
+# this alias, --status complete filters every finished record out and the
+# user sees an empty list. Map at filter time so the on-disk corpus is the
+# source of truth and the CLI surface stays the friendlier word. The other
+# choices ("running", "cancelled", "error") are stored verbatim, so they
+# fall through unchanged.
+_STATUS_CLI_TO_STORED = {"complete": "done"}
+
+
+def _status_matches(stored, requested: str) -> bool:
+    stored = (stored or "")
+    if not isinstance(stored, str):
+        stored = ""
+    target = _STATUS_CLI_TO_STORED.get(requested, requested)
+    return stored == target
+
 
 def _load_path(path: Path) -> dict | None:
     try:
@@ -72,7 +90,7 @@ def cmd_list(args):
         data = _load_path(path)
         if data is None:
             continue
-        if args.status and (data.get("status") or "") != args.status:
+        if args.status and not _status_matches(data.get("status"), args.status):
             continue
         out.append(_summarize(rp_id, data))
     out.sort(key=lambda r: r.get("started_at") or "", reverse=True)

services/docs/service.py

@@ -5,6 +5,7 @@ from dataclasses import dataclass
 from typing import List, Dict, Any
 
 from src.rag_manager import RAGManager
+from src.constants import CHROMA_DIR
 
 
 @dataclass
@@ -34,7 +35,7 @@ class DocsService:
         results = await service.query("what is async await?")
     """
 
-    def __init__(self, persist_dir: str = "data/chroma"):
+    def __init__(self, persist_dir: str = CHROMA_DIR):
         self.rag = RAGManager(persist_directory=persist_dir)
 
     async def query(self, query: str, top_k: int = 5) -> List[DocChunk]:

services/hwfit/data/hf_models.json

@@ -14036,6 +14036,29 @@
    "vision"
   ]
  },
+ {
+  "name": "google/gemma-4-12B",
+  "provider": "Google",
+  "parameter_count": "12.0B",
+  "parameters_raw": 12000000000,
+  "min_ram_gb": 24.0,
+  "recommended_ram_gb": 32.0,
+  "min_vram_gb": 24.0,
+  "quantization": "BF16",
+  "context_length": 131072,
+  "use_case": "General purpose, multimodal",
+  "is_moe": false,
+  "num_experts": null,
+  "active_experts": null,
+  "active_parameters": null,
+  "architecture": "gemma4",
+  "pipeline_tag": "image-text-to-text",
+  "release_date": "2026-04-01",
+  "gguf_sources": [],
+  "capabilities": [
+   "vision"
+  ]
+ },
  {
   "name": "google/gemma-4-31B-it",
   "provider": "Google",
@@ -19121,4 +19144,4 @@
   ],
   "_discovered": true
  }
-]
+]