feat: add ChatGPT Subscription provider (#2876)

* feat: Add ChatGPT Subscription support and related features

- Introduced a new provider option for ChatGPT Subscription in the endpoint selection UI.
- Implemented OAuth flow for ChatGPT Subscription sign-in, including polling for authorization status.
- Updated admin interface to handle ChatGPT Subscription, including disabling API key input and providing user guidance.
- Enhanced cost tracking logic to differentiate between subscription and non-subscription endpoints.
- Added new slash commands for managing skills, including listing, searching, and invoking skills.
- Implemented caching for skill catalog to optimize performance.
- Updated tests to cover new ChatGPT Subscription functionality and ensure proper endpoint probing.
- Refactored existing code to accommodate new features and improve maintainability.

* refactor: share provider device-flow setup

- reuse one device-flow backend for Copilot and ChatGPT Subscription
- add one frontend device-flow helper for Settings and /setup
- put GitHub Copilot back into Add Models, now as a dropdown option
- make provider selection just select; clicking Add starts sign-in
- stop ChatGPT Subscription setup from opening auth tabs automatically
- make /setup copilot and /setup chatgpt-subscription work from chat
- show ChatGPT Subscription in the /setup suggestions
- show the real error message when setup fails
- add focused tests for the shared flow and setup UI

* feat(chatgpt-subscription): harden credential lifecycle and streamline auth UX

Backend:
- Resolve runtime bearer for provider-auth endpoints at probe time via a
  shared _resolve_probe_key() that delegates to resolve_endpoint_runtime,
  applied across all probe/refresh call sites.
- Skip live completion probes and health pings for discovery-only providers
  (centralized behind _is_discovery_only_provider) — the Codex/Responses API
  has no such endpoints, so status is derived from cached models.
- Never persist the short lived ChatGPT bearer to the plaintext sessions
  table; proactively clear any stale bearer left by an earlier code path.
- Revoke orphaned ProviderAuthSession credentials when the last endpoint
  backing them is deleted (_delete_orphaned_provider_auth), surfaced via
  cleared_provider_auth in the delete response.

Frontend (admin.js):
- Auto-start the device-auth flow on provider selection so the authorization
  panel (code + Authorize) shows immediately instead of behind a "Sign in" click.
- Remove the redundant top button for device auth providers, move retry
  into the panel via an inline "Try again".
- Drop the self-evident hint text and add an execCommand clipboard fallback so
  Copy works in non-secure (HTTP/LAN) contexts.

* fix: harden chatgpt subscription provider

* chore: remove PR media from branch

* Fix chatgpt subscription recovery and token handling

---------

Co-authored-by: 5p00kyy <admin@5p00ky.dev>

routes/chat_helpers.py

@@ -196,14 +196,26 @@ 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()
 
@@ -212,26 +224,33 @@ 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
@@ -243,7 +262,7 @@ 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:
@@ -336,16 +355,26 @@ 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 ""
@@ -361,10 +390,30 @@ 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)
@@ -408,7 +457,12 @@ 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:
@@ -542,7 +596,11 @@ 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
 

routes/chat_routes.py

@@ -169,13 +169,20 @@ 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
@@ -194,16 +201,51 @@ 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]
@@ -213,14 +255,17 @@ 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

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/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/model_routes.py

@@ -283,6 +283,7 @@ _HOST_TO_CURATED = (
     ("fireworks.ai", "fireworks"),
     ("googleapis.com", "google"),
     ("x.ai", "xai"),
+
     ("openrouter.ai", "openrouter"),
     ("ollama.com", "ollama"),
     ("opencode.ai/zen/go", "opencode-go"),
@@ -493,6 +494,8 @@ _NON_CHAT_EXACT_PREFIXES = (
 def _is_chat_model(model_id: str) -> bool:
     """Return True if the model ID looks like a chat/completions-capable model."""
     mid = model_id.lower()
+    if mid in {"gpt-5.1-codex"}:
+        return True
     for prefix in _NON_CHAT_PREFIXES:
         if mid.startswith(prefix):
             return False
@@ -505,9 +508,67 @@ def _is_chat_model(model_id: str) -> bool:
     return True
 
 
-def _probe_single_model(base: str, api_key: str, model_id: str, timeout: int = 10, with_tools: bool = False) -> dict:
+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.
+
+    Subscription providers (e.g. ChatGPT Subscription) keep their refresh token
+    in ProviderAuthSession rather than ModelEndpoint.api_key. When the last
+    endpoint backed by that auth row is removed, the stored credentials should
+    be cleared instead of lingering. Returns True if a row was deleted.
+    ``exclude_ep_id`` drops the endpoint currently being deleted from the
+    reference count so it does not keep its own auth alive.
+    """
+    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 _is_discovery_only_provider(provider: str) -> bool:
+    """Provider that only supports model discovery, not live probing.
+
+    ChatGPT Subscription speaks the Responses/Codex API and has no
+    chat-completions or general health endpoint, so completion probes and
+    reachability pings are skipped — status is derived from cached models.
+    """
+    return provider == "chatgpt-subscription"
+
+
+def _resolve_probe_key(ep) -> Optional[str]:
+    """API key/bearer to probe an endpoint with.
+
+    Delegates to ``resolve_endpoint_runtime``, which already returns the static
+    ``ModelEndpoint.api_key`` for keyed endpoints and resolves (and refreshes)
+    the runtime bearer for session-backed providers (e.g. ChatGPT Subscription).
+    Returns None if resolution fails (e.g. re-auth required) so probing skips
+    rather than raising. Reads only already-loaded scalar attributes of ``ep``.
+    """
+    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 e:
+        logger.warning("Probe key resolution failed for %s: %s", getattr(ep, "id", "?"), e)
+        return None
+
+
+def _probe_single_model(base: str, api_key: Optional[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)
+    if _is_discovery_only_provider(provider):
+        # Responses/Codex API, not chat-completions: a completion probe would
+        # 400 and the re-probe flow would then hide every model. Discovery-only.
+        return {"status": "ok", "latency_ms": 0, "skipped": True}
     messages = [
         {"role": "system", "content": "You are a helpful assistant."},
         {"role": "user", "content": "Say OK"},
@@ -621,6 +682,11 @@ 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) == "chatgpt-subscription":
+        from src.chatgpt_subscription import fetch_available_models
+        if api_key:
+            return fetch_available_models(api_key, timeout=timeout)
+        return []
     if _detect_provider(base) == "anthropic":
         # Try Anthropic's /v1/models endpoint first
         url = build_models_url(base)
@@ -647,6 +713,10 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
             logger.warning(f"Anthropic /v1/models failed, using hardcoded list: {e}")
         return list(ANTHROPIC_MODELS)
     url = build_models_url(base)
+    if not url:
+        curated_key = _match_provider_curated(base, None)
+        fallback = _PROVIDER_CURATED.get(curated_key) if curated_key else None
+        return list(fallback or [])
     headers = build_headers(api_key, base)
     try:
         r = httpx.get(url, headers=headers, timeout=timeout, verify=llm_verify())
@@ -998,6 +1068,17 @@ def setup_model_routes(model_discovery):
                         ok, info = _should_refresh_endpoint(ep, now, force=force)
                         if not ok:
                             continue
+                        if getattr(ep, "provider_auth_id", None):
+                            try:
+                                from src.endpoint_resolver import resolve_endpoint_runtime
+                                info["base"], info["api_key"] = resolve_endpoint_runtime(
+                                    ep,
+                                    owner=getattr(ep, "owner", None),
+                                )
+                                info["key"] = _refresh_key(info["base"], info["api_key"])
+                            except Exception as e:
+                                logger.warning("Skipping model refresh for %s: could not resolve provider auth: %s", getattr(ep, "name", ep.id), e)
+                                continue
                         groups.setdefault(info["key"], {
                             "base": info["base"],
                             "api_key": info["api_key"],
@@ -1266,12 +1347,20 @@ def setup_model_routes(model_discovery):
                 "endpoint_kind": kind,
             }
             try:
-                t0 = _time.time()
-                ping = _ping_endpoint(base, ep.api_key, timeout=1.5)
-                entry["latency_ms"] = round((_time.time() - t0) * 1000)
-                entry["status"] = "online" if ping.get("reachable") or cached_count else "offline"
-                entry["error"] = ping.get("error")
-                entry["model_count"] = cached_count or (len(ANTHROPIC_MODELS) if provider == "anthropic" else 0)
+                if _is_discovery_only_provider(provider):
+                    # No general health endpoint — an unauthenticated GET just
+                    # 401s. Report status from cached models instead of pinging.
+                    entry["latency_ms"] = None
+                    entry["status"] = "online" if cached_count else "offline"
+                    entry["error"] = None
+                    entry["model_count"] = cached_count
+                else:
+                    t0 = _time.time()
+                    ping = _ping_endpoint(base, ep.api_key, timeout=1.5)
+                    entry["latency_ms"] = round((_time.time() - t0) * 1000)
+                    entry["status"] = "online" if ping.get("reachable") or cached_count else "offline"
+                    entry["error"] = ping.get("error")
+                    entry["model_count"] = cached_count or (len(ANTHROPIC_MODELS) if provider == "anthropic" else 0)
             except Exception as e:
                 entry["latency_ms"] = None
                 entry["status"] = "online" if cached_count else "offline"
@@ -1304,7 +1393,7 @@ def setup_model_routes(model_discovery):
                 if ep_id and ep_id not in endpoints_cache:
                     ep = db.query(ModelEndpoint).filter(ModelEndpoint.id == ep_id).first()
                     if ep:
-                        endpoints_cache[ep_id] = {"base_url": ep.base_url, "api_key": ep.api_key}
+                        endpoints_cache[ep_id] = {"base_url": ep.base_url, "api_key": _resolve_probe_key(ep)}
                 ep_data = endpoints_cache.get(ep_id)
                 if not ep_data:
                     # Try to find by base_url from the model's endpoint field
@@ -1343,7 +1432,7 @@ def setup_model_routes(model_discovery):
                     "id": ep.id,
                     "name": ep.name,
                     "base_url": ep.base_url,
-                    "api_key": ep.api_key,
+                    "api_key": _resolve_probe_key(ep),
                 })
         finally:
             db.close()
@@ -1432,12 +1521,14 @@ def setup_model_routes(model_discovery):
                 # Endpoint counts as reachable if it has any model — including
                 # admin-pinned IDs that a probe would never surface.
                 status = "online" if (all_models or pinned) else "offline"
+                base = _normalize_base(r.base_url)
                 ping = None
-                if not all_models and not pinned and r.is_enabled:
+                # Discovery-only providers have no health endpoint — an
+                # unauthenticated ping just 401s, so don't bother.
+                if not all_models and not pinned and r.is_enabled and not _is_discovery_only_provider(_detect_provider(base)):
                     ping = _ping_endpoint(r.base_url, r.api_key, timeout=1.0)
                     if ping.get("reachable"):
                         status = "empty"
-                base = _normalize_base(r.base_url)
                 kind = _effective_endpoint_kind(r, base)
                 results.append({
                     "id": r.id,
@@ -1713,7 +1804,7 @@ def setup_model_routes(model_discovery):
             ep = db.query(ModelEndpoint).filter(ModelEndpoint.id == ep_id).first()
             if not ep:
                 raise HTTPException(404, "Endpoint not found")
-            ep_data = {"id": ep.id, "name": ep.name, "base_url": ep.base_url, "api_key": ep.api_key}
+            ep_data = {"id": ep.id, "name": ep.name, "base_url": ep.base_url, "api_key": _resolve_probe_key(ep)}
         finally:
             db.close()
 
@@ -1777,7 +1868,7 @@ def setup_model_routes(model_discovery):
                 category = _classify_endpoint(base, kind)
                 timeout = _manual_refresh_timeout(ep, category, refresh_timeout)
                 try:
-                    probed = _probe_endpoint(base, ep.api_key, timeout=timeout)
+                    probed = _probe_endpoint(base, _resolve_probe_key(ep), timeout=timeout)
                 except Exception as exc:
                     logger.warning("Manual model refresh failed for endpoint %s at %s: %s", ep_id, base, exc)
                     probed = []
@@ -2116,7 +2207,9 @@ 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
@@ -2126,6 +2219,7 @@ 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/research_routes.py

@@ -75,6 +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"])
 
@@ -371,7 +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
@@ -380,18 +411,10 @@ 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:
@@ -408,7 +431,6 @@ def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
                 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
@@ -417,18 +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:

routes/session_routes.py

@@ -92,18 +92,13 @@ 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 and not _auth_disabled():
@@ -114,13 +109,13 @@ def _verify_session_owner(request: Request, session_id: str, session_manager=Non
     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")
 
@@ -372,8 +367,13 @@ 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/
@@ -387,8 +387,13 @@ 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:

routes/skills_routes.py

@@ -1109,6 +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,
@@ -1272,6 +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/webhook_routes.py

@@ -325,22 +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")