Cookbook launch and gallery upload fixes

routes/cookbook_routes.py

@@ -2446,6 +2446,17 @@ def setup_cookbook_routes() -> APIRouter:
 
             disk_tasks = on_disk.get("tasks") or [] if isinstance(on_disk, dict) else []
             incoming_tasks = data.get("tasks") if isinstance(data.get("tasks"), list) else []
+            incoming_removed = data.get("removedTasks") if isinstance(data.get("removedTasks"), dict) else {}
+            disk_removed = on_disk.get("removedTasks") if isinstance(on_disk, dict) and isinstance(on_disk.get("removedTasks"), dict) else {}
+            removed_tasks = {**disk_removed, **incoming_removed}
+            data["removedTasks"] = removed_tasks
+            removed_ids = set(removed_tasks.keys())
+            if removed_ids:
+                incoming_tasks = [
+                    t for t in incoming_tasks
+                    if not (isinstance(t, dict) and t.get("sessionId") in removed_ids)
+                ]
+                data["tasks"] = incoming_tasks
             # Anti-poisoning guard: a stale browser tab can keep POSTing a
             # download task as status='done' from before the strict-finish
             # fix landed, undoing any server-side correction. For each
@@ -2483,6 +2494,8 @@ def setup_cookbook_routes() -> APIRouter:
                 sid = t.get("sessionId")
                 if not sid or sid in incoming_ids:
                     continue  # client's version wins
+                if sid in removed_ids:
+                    continue  # intentional cross-device clear/remove
                 ts = t.get("ts") or 0
                 if isinstance(ts, (int, float)) and (now_ms - ts) <= RACE_WINDOW_MS:
                     preserved.append(t)

routes/model_routes.py

@@ -714,6 +714,16 @@ def _effective_endpoint_kind(ep: Any, base_url: str) -> str:
     return "auto"
 
 
+def _is_loading_model_response(resp: Any) -> bool:
+    if getattr(resp, "status_code", None) != 503:
+        return False
+    try:
+        body = resp.text or ""
+    except Exception:
+        body = ""
+    return "loading model" in body.lower()
+
+
 
 def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> List[str]:
     """Probe a base URL's /models endpoint and return list of model IDs.
@@ -778,6 +788,9 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
                         models.append(_e)
             return [m for m in models if _is_chat_model(m)]
     except httpx.HTTPStatusError as e:
+        if e.response is not None and _is_loading_model_response(e.response):
+            logger.info(f"Endpoint still loading model at {url}")
+            return []
         if api_key:
             status = e.response.status_code if e.response is not None else "unknown"
             logger.warning(f"Failed to probe {url} with API key: HTTP {status}")
@@ -827,6 +840,15 @@ def _ping_endpoint(base_url: str, api_key: str = None, timeout: float = 1.5) ->
         or "ollama" in (parsed_base.hostname or "").lower()
     )
 
+    def _is_loading_model_response(r) -> bool:
+        if getattr(r, "status_code", None) != 503:
+            return False
+        try:
+            body = r.text or ""
+        except Exception:
+            body = ""
+        return "loading model" in body.lower()
+
     def _result_from_response(r) -> Dict[str, Any]:
         if 300 <= r.status_code < 400:
             loc = r.headers.get("location", "")
@@ -843,6 +865,13 @@ def _ping_endpoint(base_url: str, api_key: str = None, timeout: float = 1.5) ->
                 "status_code": r.status_code,
                 "error": None,
             }
+        if _is_loading_model_response(r):
+            return {
+                "reachable": True,
+                "loading": True,
+                "status_code": r.status_code,
+                "error": "Loading model",
+            }
         return {"reachable": False, "status_code": r.status_code, "error": f"HTTP {r.status_code}"}
 
     last_error: Optional[str] = None
@@ -1427,7 +1456,7 @@ def setup_model_routes(model_discovery):
                 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["status"] = "loading" if ping.get("loading") else ("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:
@@ -1606,7 +1635,32 @@ def setup_model_routes(model_discovery):
                     ping_timeout = 10.0 if _classify_endpoint(base_for_ping, kind_for_ping) == "local" else 3.5
                     ping = _ping_endpoint(r.base_url, r.api_key, timeout=ping_timeout)
                     if ping.get("reachable"):
-                        status = "empty"
+                        status = "loading" if ping.get("loading") else "empty"
+                        if ping.get("loading"):
+                            base = _normalize_base(r.base_url)
+                            kind = _effective_endpoint_kind(r, base)
+                            results.append({
+                                "id": r.id,
+                                "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,
+                                "hidden_count": len(hidden),
+                                "online": True,
+                                "status": status,
+                                "ping_error": (ping or {}).get("error") if ping else None,
+                                "model_type": getattr(r, "model_type", None) or "llm",
+                                "supports_tools": getattr(r, "supports_tools", None),
+                                "endpoint_kind": kind,
+                                "category": _classify_endpoint(base, kind),
+                                "model_refresh_mode": _endpoint_refresh_mode(r, kind),
+                                "model_refresh_interval": getattr(r, "model_refresh_interval", None),
+                                "model_refresh_timeout": getattr(r, "model_refresh_timeout", None),
+                            })
+                            continue
                         # 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".
@@ -1859,7 +1913,7 @@ def setup_model_routes(model_discovery):
             "models": _merge_model_ids(model_ids, _pinned),
             "pinned_models": _pinned,
             "online": bool(model_ids) or bool(_pinned) or bool(ping.get("reachable")),
-            "status": "online" if (model_ids or _pinned) else ("empty" if ping.get("reachable") else "offline"),
+            "status": "online" if (model_ids or _pinned) else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline")),
             "ping_error": ping.get("error") if ping else None,
             "endpoint_kind": requested_kind,
             "category": _classify_endpoint(base_url, requested_kind),
@@ -1888,7 +1942,7 @@ def setup_model_routes(model_discovery):
         return {
             "base_url": base_url,
             "online": bool(models) or bool(ping.get("reachable")),
-            "status": "online" if models else ("empty" if ping.get("reachable") else "offline"),
+            "status": "online" if models else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline")),
             "ping_error": ping.get("error") if ping else None,
             "models": models,
             "count": len(models),

routes/shell_routes.py

@@ -1108,7 +1108,7 @@ def setup_shell_routes() -> APIRouter:
             {
                 "name": "llama_cpp",
                 "pip": "llama-cpp-python[server]",
-                "desc": "Serve GGUF models via llama.cpp",
+                "desc": "Great for single-GPU or CPU inference with GGUF models",
                 "category": "LLM",
                 "target": "remote",
                 # Build-toolchain prereqs. Cookbook's launch bootstrap
@@ -1129,7 +1129,7 @@ def setup_shell_routes() -> APIRouter:
             {
                 "name": "vllm",
                 "pip": "vllm",
-                "desc": "High-throughput LLM serving engine",
+                "desc": "Great for high-throughput multi-GPU inference",
                 "category": "LLM",
                 "target": "remote",
             },

routes/upload_routes.py

@@ -3,11 +3,16 @@ import os
 import time
 import json
 import asyncio
+import shutil
+import uuid
+from pathlib import Path
 from fastapi import APIRouter, Request, File, UploadFile, HTTPException
 from typing import List
 import logging
 from core.middleware import require_admin
+from core.database import SessionLocal, GalleryImage
 from src.auth_helpers import effective_user
+from src.constants import GENERATED_IMAGES_DIR
 from src.upload_handler import count_recent_uploads
 
 logger = logging.getLogger(__name__)
@@ -50,6 +55,69 @@ def setup_upload_routes(upload_handler):
             raise HTTPException(404, "File not found")
 
         raise HTTPException(404, "File not found")
+
+    def _promote_chat_image_to_gallery(meta: dict, owner: str | None) -> str | None:
+        """Make chat-uploaded images visible in Gallery without changing chat storage."""
+        is_image_file = getattr(upload_handler, "is_image_file", None)
+        if not callable(is_image_file):
+            return None
+        if not is_image_file(meta.get("name", ""), meta.get("mime", "")):
+            return None
+
+        source_path = meta.get("path")
+        if not source_path or not os.path.isfile(source_path):
+            return None
+
+        db = SessionLocal()
+        try:
+            file_hash = meta.get("hash")
+            if file_hash:
+                q = db.query(GalleryImage).filter(
+                    GalleryImage.file_hash == file_hash,
+                    GalleryImage.is_active == True,  # noqa: E712
+                )
+                if owner:
+                    q = q.filter(GalleryImage.owner == owner)
+                existing = q.first()
+                if existing:
+                    return existing.id
+
+            image_dir = Path(GENERATED_IMAGES_DIR)
+            image_dir.mkdir(parents=True, exist_ok=True)
+            ext = Path(meta.get("name") or source_path).suffix.lower()
+            if ext not in {".png", ".jpg", ".jpeg", ".webp", ".gif"}:
+                mime_ext = {
+                    "image/png": ".png",
+                    "image/jpeg": ".jpg",
+                    "image/jpg": ".jpg",
+                    "image/webp": ".webp",
+                    "image/gif": ".gif",
+                }.get(meta.get("mime", ""))
+                ext = mime_ext or ".png"
+            filename = f"{uuid.uuid4().hex[:12]}{ext}"
+            dest_path = image_dir / filename
+            shutil.copy2(source_path, dest_path)
+
+            image_id = str(uuid.uuid4())
+            db.add(GalleryImage(
+                id=image_id,
+                filename=filename,
+                prompt=meta.get("name") or "Chat upload",
+                model="chat-upload",
+                owner=owner,
+                file_hash=file_hash,
+                width=meta.get("width"),
+                height=meta.get("height"),
+                file_size=meta.get("size"),
+            ))
+            db.commit()
+            return image_id
+        except Exception as e:
+            db.rollback()
+            logger.warning("Failed to add chat image upload to gallery: %s", e)
+            return None
+        finally:
+            db.close()
     
     @router.post("")
     async def api_upload(request: Request, files: List[UploadFile] = File(...)):
@@ -78,8 +146,10 @@ def setup_upload_routes(upload_handler):
         
         for u in files:
             try:
-                meta = upload_handler.save_upload(u, client_ip, owner=effective_user(request))
+                owner = effective_user(request)
-                out.append({
+                meta = upload_handler.save_upload(u, client_ip, owner=owner)
+                gallery_id = _promote_chat_image_to_gallery(meta, owner)
+                item = {
                     "id": meta["id"],
                     "name": meta["name"],
                     "mime": meta["mime"],
@@ -89,7 +159,10 @@ def setup_upload_routes(upload_handler):
                     "width": meta.get("width"),
                     "height": meta.get("height"),
                     "is_duplicate": meta.get("is_duplicate", False)
-                })
+                }
+                if gallery_id:
+                    item["gallery_id"] = gallery_id
+                out.append(item)
             except HTTPException:
                 raise
             except Exception as e:

src/llm_core.py

@@ -907,7 +907,10 @@ def _anthropic_rejects_temperature(model: str) -> bool:
     return (int(match.group(1)), int(match.group(2))) >= (4, 7)
 
 # Models that support structured thinking — may output </think> without opening tag
-_THINKING_MODEL_PATTERNS = ("qwen3", "qwq", "deepseek-r1", "deepseek-reasoner", "minimax", "m2-reap", "gemma")
+_THINKING_MODEL_PATTERNS = (
+    "qwen3", "qwq", "deepseek-r1", "deepseek-reasoner", "minimax",
+    "m2-reap", "gemma", "stepfun", "step-3", "step3",
+)
 
 def _supports_thinking(model: str) -> bool:
     """Check if model supports structured thinking output."""

src/tool_parsing.py

@@ -39,6 +39,10 @@ _XML_TOOL_CALL_RE = re.compile(
     r"<(?:[\w]+:)?(?:tool_call|function_call)>\s*([\s\S]*?)</(?:[\w]+:)?(?:tool_call|function_call)>",
     re.IGNORECASE,
 )
+_XML_OPEN_TOOL_CALL_RE = re.compile(
+    r"<(?:[\w]+:)?(?:tool_call|function_call)>\s*([\s\S]*)\Z",
+    re.IGNORECASE,
+)
 _XML_INVOKE_RE = re.compile(
     r'<invoke\s+name=["\'](\w+)["\']>\s*([\s\S]*?)</invoke>',
     re.IGNORECASE,
@@ -47,6 +51,24 @@ _XML_PARAM_RE = re.compile(
     r'<parameter\s+name=["\'](\w+)["\']>([\s\S]*?)</parameter>',
     re.IGNORECASE,
 )
+_XML_DIRECT_TOOL_RE = re.compile(
+    r"<\s*([A-Za-z_][\w-]*)\s*>([\s\S]*?)</\s*\1\s*>",
+    re.IGNORECASE,
+)
+
+# Pattern 3b: StepFun Step-3.x native tool-call tokens. The tokenizer defines:
+#   <｜tool▁calls▁begin｜> ... <｜tool▁calls▁end｜>
+#   <｜tool▁call▁begin｜>tool_name<｜tool▁sep｜>{...}<｜tool▁call▁end｜>
+# These can leak as text through llama.cpp/Ollama-style endpoints when the
+# engine does not return structured OpenAI tool_calls.
+_STEPFUN_TOOL_CALL_RE = re.compile(
+    r"<｜tool▁call▁begin｜>\s*([A-Za-z_][\w.-]*)\s*<｜tool▁sep｜>\s*([\s\S]*?)\s*<｜tool▁call▁end｜>",
+    re.IGNORECASE,
+)
+_STEPFUN_TOOL_CALLS_WRAPPER_RE = re.compile(
+    r"</?｜tool▁calls▁(?:begin|end)｜>",
+    re.IGNORECASE,
+)
 
 # Pattern 4: <tool_code> blocks (MiniMax-M2.5 style)
 # {tool => 'tool_name', args => '<param>value</param>'}
@@ -446,6 +468,76 @@ def _parse_xml_invoke(inv_match) -> Optional[ToolBlock]:
     return function_call_to_tool_block(tool_name, json.dumps(params))
 
 
+def _parse_xml_direct_tool(tool_match) -> Optional[ToolBlock]:
+    """Parse direct XML tool tags inside <tool_call>.
+
+    Some local models emit:
+      <tool_call><web_search>query</web_search></tool_call>
+    instead of the invoke/parameter shape:
+      <tool_call><invoke name="web_search"><parameter name="query">query</parameter></invoke></tool_call>
+    Keep this as an adapter to the canonical function-call converter so aliases
+    and per-tool argument formatting stay in one place.
+    """
+    tool_name = tool_match.group(1).lower().replace("-", "_")
+    if tool_name in {"invoke", "parameter", "tool_call", "function_call"}:
+        return None
+    mapped = _TOOL_NAME_MAP.get(tool_name) or (tool_name if tool_name in TOOL_TAGS else None)
+    if not mapped:
+        return None
+    body = tool_match.group(2).strip()
+    if not body:
+        return None
+    try:
+        params = json.loads(body)
+        if not isinstance(params, dict):
+            params = {}
+    except json.JSONDecodeError:
+        if mapped == "web_search":
+            params = {"query": body}
+        elif mapped == "web_fetch":
+            params = {"url": body}
+        elif mapped == "bash":
+            params = {"command": body}
+        elif mapped == "python":
+            params = {"code": body}
+        elif mapped in ("read_file", "write_file"):
+            params = {"path": body}
+        else:
+            params = {"content": body}
+    from src.tool_schemas import function_call_to_tool_block
+    return function_call_to_tool_block(mapped, json.dumps(params))
+
+
+def _parse_stepfun_tool_call(call_match) -> Optional[ToolBlock]:
+    """Parse StepFun native tool-call tokens into an Odysseus ToolBlock."""
+    tool_name = call_match.group(1).lower().replace("-", "_").replace(".", "_")
+    mapped = _TOOL_NAME_MAP.get(tool_name) or (tool_name if tool_name in TOOL_TAGS else None)
+    if not mapped:
+        return None
+    body = call_match.group(2).strip()
+    if not body:
+        return None
+    try:
+        params = json.loads(body)
+        if not isinstance(params, dict):
+            params = {}
+    except json.JSONDecodeError:
+        if mapped == "web_search":
+            params = {"query": body}
+        elif mapped == "web_fetch":
+            params = {"url": body}
+        elif mapped == "bash":
+            params = {"command": body}
+        elif mapped == "python":
+            params = {"code": body}
+        elif mapped in ("read_file", "write_file"):
+            params = {"path": body}
+        else:
+            params = {"content": body}
+    from src.tool_schemas import function_call_to_tool_block
+    return function_call_to_tool_block(mapped, json.dumps(params))
+
+
 def _parse_tool_code_block(raw: str) -> Optional[ToolBlock]:
     """Parse a <tool_code>{tool => 'name', args => '...'}</tool_code> block (MiniMax style)."""
     # Extract tool name
@@ -511,8 +603,9 @@ def parse_tool_blocks(text: str, skip_fenced: bool = False) -> List[ToolBlock]:
     2. [TOOL_CALL] ... [/TOOL_CALL] blocks (some models)
     3. XML-style <tool_call>/<invoke> blocks
     4. <tool_code> blocks (MiniMax-M2.5 style)
-    5. DeepSeek DSML markup (normalized to <invoke> first)
+    5. StepFun Step-3 native <｜tool▁call▁begin｜> tokens
-    6. Non-native local model fallback: prose mentioning web_search followed by
+    6. DeepSeek DSML markup (normalized to <invoke> first)
+    7. Non-native local model fallback: prose mentioning web_search followed by
        bare JSON args, e.g. {"query":"...", "time_filter":"week"}
 
     `skip_fenced`: when True, Pattern 1 (fenced ```bash/```python/```json code
@@ -567,12 +660,38 @@ def parse_tool_blocks(text: str, skip_fenced: bool = False) -> List[ToolBlock]:
 
     # Pattern 3: XML-style <tool_call>/<invoke> blocks
     if not blocks:
+        for step_call in _STEPFUN_TOOL_CALL_RE.finditer(text):
+            block = _parse_stepfun_tool_call(step_call)
+            if block:
+                blocks.append(block)
+        if blocks:
+            return blocks
         # Try wrapped: <tool_call><invoke ...>...</invoke></tool_call>
         for m in _XML_TOOL_CALL_RE.finditer(text):
             for inv in _XML_INVOKE_RE.finditer(m.group(1)):
                 block = _parse_xml_invoke(inv)
                 if block:
                     blocks.append(block)
+            if not blocks:
+                for direct in _XML_DIRECT_TOOL_RE.finditer(m.group(1)):
+                    block = _parse_xml_direct_tool(direct)
+                    if block:
+                        blocks.append(block)
+        # Some local models stream an opening <tool_call> wrapper and a
+        # complete inner tool tag, but forget the closing </tool_call>.
+        if not blocks:
+            for m in _XML_OPEN_TOOL_CALL_RE.finditer(text):
+                body = m.group(1)
+                for inv in _XML_INVOKE_RE.finditer(body):
+                    block = _parse_xml_invoke(inv)
+                    if block:
+                        blocks.append(block)
+                if blocks:
+                    break
+                for direct in _XML_DIRECT_TOOL_RE.finditer(body):
+                    block = _parse_xml_direct_tool(direct)
+                    if block:
+                        blocks.append(block)
         # Try bare <invoke> without wrapper
         if not blocks:
             for inv in _XML_INVOKE_RE.finditer(text):
@@ -614,7 +733,10 @@ def strip_tool_blocks(text: str, skip_fenced: bool = False) -> str:
     text = _normalize_dsml(text)
     cleaned = text if skip_fenced else _TOOL_BLOCK_RE.sub('', text)
     cleaned = _TOOL_CALL_RE.sub('', cleaned)
+    cleaned = _STEPFUN_TOOL_CALL_RE.sub('', cleaned)
+    cleaned = _STEPFUN_TOOL_CALLS_WRAPPER_RE.sub('', cleaned)
     cleaned = _XML_TOOL_CALL_RE.sub('', cleaned)
+    cleaned = _XML_OPEN_TOOL_CALL_RE.sub('', cleaned)
     cleaned = _TOOL_CODE_RE.sub('', cleaned)
     if not skip_fenced:
         raw_web_json = _parse_raw_web_json_lookup(cleaned)

static/js/cookbook.js

@@ -230,14 +230,30 @@ export function _isMetal() {
 }
 
 /** Detect model-specific vLLM optimizations */
+function _isStepFunStepModel(modelName) {
+  const n = (modelName || '').toLowerCase();
+  return n.includes('stepfun')
+    || n.includes('step-3')
+    || n.includes('step3')
+    || n.includes('step_3');
+}
+
 function _detectModelOptimizations(modelName) {
   const n = (modelName || '').toLowerCase();
   const opts = { envVars: [], flags: [], tips: [] };
 
+  // StepFun Step-3.x MoE models. Their tokenizer defines the Step tool-call
+  // and thinking tags; vLLM/SGLang need the step3p5 parser instead of generic
+  // Hermes/XML guesses, and the MoE backend should default to expert parallel.
+  if (_isStepFunStepModel(modelName)) {
+    opts.flags.push('--enable-expert-parallel');
+    opts.tips.push('StepFun Step-3 MoE: expert parallel');
+    opts.tips.push('StepFun parser: step3p5 for native tool calls and reasoning tags');
+  }
   // Qwen3.5 MoE models — MoE-specific env vars + expert-parallel.
   // The --reasoning-parser flag is added uniformly below via
   // _detectReasoningParser, no longer hardcoded here.
-  if (n.includes('qwen3.5') || n.includes('qwen3-') && (n.includes('a10b') || n.includes('a22b') || n.includes('a3b'))) {
+  else if (n.includes('qwen3.5') || n.includes('qwen3-') && (n.includes('a10b') || n.includes('a22b') || n.includes('a3b'))) {
     opts.envVars.push('VLLM_USE_DEEP_GEMM=0', 'VLLM_USE_FLASHINFER_MOE_FP16=1', 'VLLM_USE_FLASHINFER_SAMPLER=0', 'OMP_NUM_THREADS=4');
     opts.flags.push('--enable-expert-parallel');
     opts.tips.push('MoE optimizations: expert parallel + flashinfer MoE kernels');
@@ -312,6 +328,9 @@ function _detectModelOptimizations(modelName) {
  */
 export function _detectReasoningParser(modelName) {
   const n = (modelName || '').toLowerCase();
+  // StepFun Step-3.x uses Step's native <think> / tool-call tokens. vLLM
+  // registers this parser as step3p5.
+  if (_isStepFunStepModel(modelName)) return 'step3p5';
   // MiniMax M3 — newer vLLM nightly/parser builds use minimax_m3. This must
   // be checked before the M2.x rule and before the generic MiniMax tool parser.
   if (n.includes('minimax') && /\bm3\b/.test(n)) return 'minimax_m3';
@@ -348,6 +367,7 @@ export function _detectReasoningParser(modelName) {
  */
 export function _detectToolParser(modelName) {
   const n = (modelName || '').toLowerCase();
+  if (_isStepFunStepModel(modelName)) return 'step3p5';
   if (n.includes('qwen3') && n.includes('coder')) return 'qwen3_coder';
   if (n.includes('qwen3')) return 'qwen3_xml';
   if (n.includes('qwen')) return 'hermes';   // Qwen2.5 / Qwen2 / Qwen1.5
@@ -601,6 +621,13 @@ export function _buildServeCmd(f, modelName, backend) {
     if (f.dtype && f.dtype !== 'auto') cmd += ` --dtype ${f.dtype}`;
     if (f.max_seqs && f.max_seqs.toString().trim()) cmd += ` --max-running-requests ${f.max_seqs.toString().trim()}`;
     if (f.trust_remote) cmd += ' --trust-remote-code';
+    if (f.auto_tool) cmd += ` --enable-auto-tool-choice --tool-call-parser ${_detectToolParser(modelName)}`;
+    if (f.expert_parallel) cmd += ' --enable-expert-parallel';
+    if (f.reasoning_parser) {
+      const rp = typeof f.reasoning_parser === 'string' && f.reasoning_parser !== 'true'
+        ? f.reasoning_parser : (f._reasoning_parser_value || _detectReasoningParser(modelName) || '');
+      if (rp) cmd += ` --reasoning-parser ${rp}`;
+    }
     if (!f.prefix_cache) cmd += ' --disable-radix-cache';
     if (f.enforce_eager) cmd += ' --disable-cuda-graph';
   } else if (backend === 'llamacpp') {
@@ -909,10 +936,10 @@ async function _fetchDependencies() {
     // matches the engine you're configuring. Unknown packages get no
     // icon (the name alone is fine for librosa, hf_transfer, etc.).
     const _DEP_GLYPHS = {
-      vllm:    '<svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.4" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M3 4l7 16 7-16"/><path d="M14 4l4 9 3-9"/></svg>',
+      vllm: '<svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.4" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M3 4l7 16 7-16"/><path d="M14 4l4 9 3-9"/></svg>',
-      sglang:  '<svg width="13" height="13" viewBox="0 0 24 24" fill="currentColor" stroke="none" aria-hidden="true"><polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"/></svg>',
+      sglang: '<span aria-hidden="true" style="display:block;width:13px;height:13px;background:currentColor;-webkit-mask:url(/static/icons/sglang-mark.png) center/contain no-repeat;mask:url(/static/icons/sglang-mark.png) center/contain no-repeat;"></span>',
-      llama_cpp: '<svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><circle cx="12" cy="12" r="9"/><path d="M8 12h8M12 8v8"/></svg>',
+      llama_cpp: '<svg width="13" height="13" viewBox="0 0 600 600" fill="none" aria-hidden="true"><path d="M600 392L504.249 558L504.137 557.929C487.252 584.069 458.193 600 426.864 600H120L240 392H600Z" fill="currentColor"/><path d="M240 392H0L199.602 46.0254C216.032 17.5463 246.411 0 279.29 0H466.154L240 392Z" fill="currentColor"/></svg>',
-      ollama:  '<svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M6 10a6 6 0 0 1 12 0v4a4 4 0 0 1-8 0v-1"/><circle cx="10" cy="9" r="1"/><circle cx="14" cy="9" r="1"/></svg>',
+      ollama: '<img src="/static/icons/ollama-mark-crop.png" alt="" aria-hidden="true" width="13" height="13" style="display:block;width:13px;height:13px;object-fit:contain;" />',
       diffusers: '<svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><circle cx="12" cy="12" r="4"/><path d="M12 2v3M12 19v3M2 12h3M19 12h3M5 5l2 2M17 17l2 2M5 19l2-2M17 7l2-2"/></svg>',
     };
     const _depGlyphHtml = (name) => {
@@ -2138,7 +2165,10 @@ function _wireTabEvents(body) {
         }
       }
       const shortName = repo.split('/').pop();
-      _retryDownload(shortName, payload);
+      const displayName = payload.include
+        ? `${shortName} · ${_ggufQuantFromPath(String(payload.include).replace(/\*/g, '')) || String(payload.include).replace(/\*/g, '').replace(/\.gguf$/i, '')}`
+        : shortName;
+      _retryDownload(displayName, payload);
       dlInput.value = '';
     };
     dlBtn.addEventListener('click', triggerDownload);
@@ -2179,18 +2209,13 @@ function _wireTabEvents(body) {
       const folded = dlFoldBody.classList.contains('is-folded');
       _setFolded(!folded);
     });
-    // Auto-fold on any downward scroll inside the cookbook modal,
+    // Auto-fold on any downward scroll inside the cookbook modal. Do not
-    // and auto-expand when the user scrolls all the way back to the
+    // auto-expand on upward/top scroll — once the user collapses Download,
-    // top of whichever scroller they're in. The chevron ▸ still
+    // it should stay collapsed until the header is clicked again.
-    // toggles manually.
     const _maybeFold = () => {
       if (dlFoldBody.classList.contains('is-folded')) return;
       _setFolded(true, /* persist */ false);
     };
-    const _maybeExpand = () => {
-      if (!dlFoldBody.classList.contains('is-folded')) return;
-      _setFolded(false, /* persist */ false);
-    };
     // Capture phase so scrolls on nested scrollers (.hwfit-list,
     // .cookbook-body, .modal-content) all hit us.
     const _modal = dlFold.closest('#cookbook-modal') || document;
@@ -2205,7 +2230,6 @@ function _wireTabEvents(body) {
       const y = tgt.scrollTop;
       const prev = _lastY.get(tgt) || 0;
       if (y > prev) _maybeFold();
-      else if (y <= 0) _maybeExpand();
       _lastY.set(tgt, y);
     }, true);
   }
@@ -2621,10 +2645,10 @@ function _renderRecipes() {
   html += `<input type="text" class="cookbook-dl-repo" id="cookbook-dl-repo" placeholder="org/model-name, qwen2.5:14b, or HF URL" style="flex:1;min-width:0;" />`;
   html += `<button class="cookbook-btn cookbook-dl-btn" id="cookbook-dl-btn">Download</button>`;
   html += `</div>`;
-  html += `<div id="cookbook-dl-gguf-row" style="display:none;margin-top:1px;gap:5px;align-items:center;font-size:11px;">`;
+  html += `<div id="cookbook-dl-gguf-row" class="cookbook-dl-gguf-row" style="display:none;">`;
-  html += `<span style="opacity:0.65;flex-shrink:0;">GGUF</span>`;
+  html += `<span class="cookbook-dl-gguf-label">GGUF</span>`;
-  html += `<select class="cookbook-field-input" id="cookbook-dl-gguf-quant" style="height:28px;min-width:118px;flex:0 0 auto;"></select>`;
+  html += `<select class="cookbook-field-input" id="cookbook-dl-gguf-quant"></select>`;
-  html += `<span id="cookbook-dl-gguf-note" style="opacity:0.55;white-space:nowrap;overflow:hidden;text-overflow:ellipsis;"></span>`;
+  html += `<span id="cookbook-dl-gguf-note"></span>`;
   html += `</div>`;
   // Ollama-library browse used to live here as its own collapsible dropdown,
   // but that duplicated the Engine filter (which already has Ollama). The
@@ -3047,6 +3071,56 @@ export function isVisible() {
   return !modal.classList.contains('hidden');
 }
 
+let _sharedSyncInFlight = false;
+let _sharedSyncLast = 0;
+async function _refreshSharedCookbookState(reason = '') {
+  if (!isVisible() || _sharedSyncInFlight) return;
+  const now = Date.now();
+  if (now - _sharedSyncLast < 1500) return;
+  _sharedSyncInFlight = true;
+  _sharedSyncLast = now;
+  try {
+    const ok = await _syncFromServer();
+    if (!ok) return;
+    try { Object.assign(_envState, _readStoredEnvState()); } catch {}
+    const modal = document.getElementById('cookbook-modal');
+    const activeTab = modal?.querySelector('.cookbook-tab.active')?.dataset?.backend || '';
+    if (activeTab === 'Running') {
+      _renderRunningTab();
+    } else if (activeTab === 'Settings') {
+      const active = document.activeElement;
+      const editingSettings = active && active.closest && active.closest('.cookbook-settings-stack');
+      if (!editingSettings) {
+        _renderRecipes();
+        const tab = document.querySelector('#cookbook-modal .cookbook-tab[data-backend="Settings"]');
+        if (tab) tab.click();
+      }
+    }
+  } catch (e) {
+    console.warn('[cookbook] shared state refresh failed', reason, e);
+  } finally {
+    _sharedSyncInFlight = false;
+  }
+}
+
+document.addEventListener('cookbook:state-synced', () => {
+  try { Object.assign(_envState, _readStoredEnvState()); } catch {}
+  if (isVisible()) {
+    const activeTab = document.querySelector('#cookbook-modal .cookbook-tab.active')?.dataset?.backend || '';
+    if (activeTab === 'Running') _renderRunningTab();
+  }
+});
+
+window.addEventListener('focus', () => { _refreshSharedCookbookState('focus'); });
+document.addEventListener('visibilitychange', () => {
+  if (document.visibilityState === 'visible') _refreshSharedCookbookState('visible');
+});
+setInterval(() => {
+  if (!isVisible()) return;
+  const activeTab = document.querySelector('#cookbook-modal .cookbook-tab.active')?.dataset?.backend || '';
+  if (activeTab === 'Running') _refreshSharedCookbookState('active-poll');
+}, 5000);
+
 // Close button
 document.addEventListener('DOMContentLoaded', () => {
   const closeBtn = document.getElementById('close-cookbook-modal');

static/js/cookbookDownload.js

@@ -85,6 +85,22 @@ function _ggufIncludePattern(model, source) {
   return '*.gguf';
 }
 
+function _ggufDisplayPartFromInclude(include) {
+  const clean = String(include || '').replace(/\*/g, '');
+  const parts = clean.split('/').filter(Boolean);
+  const file = parts[parts.length - 1] || clean;
+  const dir = parts.length > 1 ? parts[parts.length - 2] : '';
+  const quant = `${dir} ${file}`.match(/\b(?:UD-)?(?:IQ[1-8]_[A-Z0-9]+|Q[2-8]_K_[MLS]|Q[2-8]_[0-9A-Z]+|Q[2-8])\b/i);
+  if (quant) return quant[0].toUpperCase().replace(/^UD-/, '');
+  return file.replace(/\.gguf$/i, '').replace(/-\d{5}-of-\d{5}$/i, '');
+}
+
+function _downloadTaskName(shortName, payload) {
+  const include = payload?.include || '';
+  const part = include ? _ggufDisplayPartFromInclude(include) : '';
+  return part ? `${shortName} · ${part}` : shortName;
+}
+
 function _missingGgufMessage(model) {
   const name = model?.name || 'this model';
   if (/\bnvfp4\b/i.test(name)) {
@@ -519,6 +535,7 @@ export async function _runModelDownload(panel, model, backend, hostOverride) {
   }
 
   const shortName = (model.name || repo).split('/').pop();
+  const taskName = _downloadTaskName(shortName, payload);
   const targetHost = host || 'local';
 
   const tasks = _loadTasks();
@@ -576,7 +593,7 @@ export async function _runModelDownload(panel, model, backend, hostOverride) {
   if (activeOnHost) {
     const queueId = `queue-${Date.now().toString(36)}`;
     const allTasks = _loadTasks();
-    allTasks.push({ id: queueId, sessionId: queueId, name: shortName, type: 'download', status: 'queued', output: '', ts: Date.now(), payload, remoteHost: host });
+    allTasks.push({ id: queueId, sessionId: queueId, name: taskName, type: 'download', status: 'queued', output: '', ts: Date.now(), payload, remoteHost: host });
     _saveTasks(allTasks);
     _renderRunningTab();
     uiModule.showToast(`Queued ${shortName} — waiting for current download`);
@@ -601,8 +618,8 @@ export async function _runModelDownload(panel, model, backend, hostOverride) {
       uiModule.showToast('Download failed: ' + (data.error || ''), 9000);
       return;
     }
-    _addTask(data.session_id, shortName, 'download', payload);
+    _addTask(data.session_id, taskName, 'download', payload);
-    uiModule.showToast(`Downloading ${shortName}...`);
+    uiModule.showToast(`Downloading ${taskName}...`);
   } catch (e) {
     uiModule.showToast('Download failed: ' + e.message, 9000);
   }

static/js/cookbookRunning.js

@@ -38,6 +38,47 @@ function _taskBadge(task) {
   return { text: _statusLabel(task.status, task.type), cls: 'cookbook-task-' + task.status };
 }
 
+function _ggufDisplayPartFromPath(path) {
+  const parts = String(path || '').split('/').filter(Boolean);
+  const file = parts[parts.length - 1] || '';
+  const dir = parts.length > 1 ? parts[parts.length - 2] : '';
+  const text = `${dir} ${file}`;
+  const quant = text.match(/\b(?:UD-)?(?:IQ[1-8]_[A-Z0-9]+|Q[2-8]_K_[MLS]|Q[2-8]_[0-9A-Z]+|Q[2-8])\b/i);
+  if (quant) return quant[0].toUpperCase().replace(/^UD-/, '');
+  return file.replace(/\.gguf$/i, '').replace(/-\d{5}-of-\d{5}$/i, '');
+}
+
+function _downloadDisplayName(name, task) {
+  const include = task?.payload?.include || '';
+  if (!include || String(name || '').includes(' · ')) return name;
+  const part = _ggufDisplayPartFromPath(include.replace(/\*/g, ''));
+  return part ? `${name} · ${part}` : name;
+}
+
+function _taskDisplayName(task) {
+  const name = String(task?.name || '').trim();
+  if (task?.type === 'download') return _downloadDisplayName(name, task);
+  if (task?.type !== 'serve') return name;
+  const gguf = task?.payload?._fields?.gguf_file || task?.payload?.gguf_file || '';
+  if (!gguf || name.includes(' · ')) return name;
+  const part = _ggufDisplayPartFromPath(gguf);
+  return part ? `${name} · ${part}` : name;
+}
+
+function _canLaunchDownloadedTask(task) {
+  return task?.type === 'download' && ['done', 'completed'].includes(task.status || '') && !!(task.payload?.repo_id || task.name);
+}
+
+function _downloadServeFields(task) {
+  const include = String(task?.payload?.include || '').trim();
+  if (!include) return null;
+  return {
+    backend: 'llamacpp',
+    _forceBackend: true,
+    _preferredGgufInclude: include,
+  };
+}
+
 // A download task whose tmux output still shows an active per-shard line
 // (e.g. "model-00012-of-00082.safetensors: 56%|") is NOT actually finished —
 // the cookbook just lost track. The clear pill becomes a "reconnect" affordance
@@ -282,6 +323,40 @@ let _detectToolParser;
 let _detectModelOptimizations;
 let _buildServeCmd;
 
+function _taskServerSelection(task) {
+  const host = task?.remoteHost || task?.payload?.remote_host || '';
+  const savedKey = task?.remoteServerKey || task?.payload?.remote_server_key || '';
+  const server = (savedKey ? _serverByVal(savedKey) : null)
+    || (host ? _serverByVal(host) : null)
+    || (host ? _envState.servers.find(s => s.host === host) : null)
+    || null;
+  const key = server ? (_serverKey ? _serverKey(server) : savedKey) : (savedKey || (host || 'local'));
+  return { host, server, key };
+}
+
+function _selectTaskServer(task) {
+  const { host, server, key } = _taskServerSelection(task);
+  _envState.remoteHost = host;
+  _envState.remoteServerKey = key === 'local' ? '' : key;
+  if (server) {
+    _envState.env = server.env || 'none';
+    _envState.envPath = server.envPath || '';
+    _envState.platform = server.platform || '';
+  } else if (!host) {
+    _envState.env = 'none';
+    _envState.envPath = '';
+    _envState.platform = '';
+  }
+  document.querySelectorAll('#hwfit-server-select, #hwfit-dl-server, #hwfit-cache-server, #hwfit-deps-server').forEach(sel => {
+    if (!sel || sel.tagName !== 'SELECT') return;
+    const wanted = key || (host || 'local');
+    if ([...sel.options].some(o => o.value === wanted)) sel.value = wanted;
+    else if (host && [...sel.options].some(o => o.value === host)) sel.value = host;
+    else sel.value = host ? wanted : 'local';
+  });
+  return { host, server, key };
+}
+
 // When a new action is started (download / dependency / serve), this holds the
 // new task's id so the next render collapses every other card and leaves only
 // the new one open. Consumed (cleared) by _renderRunningTab.
@@ -654,16 +729,31 @@ function _loadPrunedTasks() {
 const _REMOVED_KEY = 'cookbook-removed-tasks';
 const _TOMBSTONE_TTL_MS = 24 * 3600 * 1000;
 function _loadTombstones() {
-  try { return JSON.parse(localStorage.getItem(_REMOVED_KEY)) || {}; }
+  try {
+    const tomb = JSON.parse(localStorage.getItem(_REMOVED_KEY)) || {};
+    const now = Date.now();
+    let changed = false;
+    for (const k in tomb) {
+      if (now - tomb[k] > _TOMBSTONE_TTL_MS) {
+        delete tomb[k];
+        changed = true;
+      }
+    }
+    if (changed) localStorage.setItem(_REMOVED_KEY, JSON.stringify(tomb));
+    return tomb;
+  }
   catch { return {}; }
 }
+function _saveTombstones(tomb) {
+  localStorage.setItem(_REMOVED_KEY, JSON.stringify(tomb || {}));
+}
 function _tombstoneTask(id) {
   if (!id) return;
   const tomb = _loadTombstones();
   const now = Date.now();
   tomb[id] = now;
   for (const k in tomb) { if (now - tomb[k] > _TOMBSTONE_TTL_MS) delete tomb[k]; }
-  localStorage.setItem(_REMOVED_KEY, JSON.stringify(tomb));
+  _saveTombstones(tomb);
 }
 function _isTombstoned(id) {
   const ts = _loadTombstones()[id];
@@ -1098,6 +1188,7 @@ function _syncToServer() {
       if (!_envState || !Array.isArray(_envState.servers) || _envState.servers.length === 0) return;
       const state = {
         tasks: _loadTasks(),
+        removedTasks: _loadTombstones(),
         presets: _loadPresets(),
         env: _envState,
         serveState: null,
@@ -1146,9 +1237,16 @@ export async function _syncFromServer() {
 
     const localTasks = _loadTasks();
     const serverTasks = state.tasks || [];
+    const serverTombstones = (state.removedTasks && typeof state.removedTasks === 'object') ? state.removedTasks : {};
+    const localTombstones = _loadTombstones();
+    const mergedTombstones = { ...serverTombstones, ...localTombstones };
+    for (const [id, ts] of Object.entries(serverTombstones)) {
+      if (localTombstones[id] == null || Number(ts) > Number(localTombstones[id])) mergedTombstones[id] = ts;
+    }
+    _saveTombstones(mergedTombstones);
 
     const localIds = new Set(localTasks.map(t => t.sessionId));
-    const merged = [...localTasks];
+    const merged = localTasks.filter(t => !_isTombstoned(t.sessionId));
     for (const t of serverTasks) {
       if (!localIds.has(t.sessionId) && !_isTombstoned(t.sessionId)) {
         merged.push(t);
@@ -1165,6 +1263,18 @@ export async function _syncFromServer() {
       const { remoteHost: _rh, env: _e, envPath: _ep, platform: _pf, ...settings } = state.env;
       delete settings.hfToken;
       Object.assign(_envState, settings);
+      const selected = (_envState.remoteServerKey && _serverByVal?.(_envState.remoteServerKey))
+        || (_envState.remoteHost ? (_envState.servers || []).find(s => s.host === _envState.remoteHost) : null);
+      if (selected) {
+        _envState.env = selected.env || 'none';
+        _envState.envPath = selected.envPath || '';
+        _envState.platform = selected.platform || '';
+      } else if (!_envState.remoteHost) {
+        const local = (_envState.servers || []).find(s => !s.host || s.host === 'local');
+        _envState.env = local?.env || 'none';
+        _envState.envPath = local?.envPath || '';
+        _envState.platform = local?.platform || '';
+      }
       const { hfToken, ...safeState } = _envState;
       localStorage.setItem('cookbook-last-state', JSON.stringify(safeState));
     }
@@ -1174,6 +1284,7 @@ export async function _syncFromServer() {
     if (state.serveState) {
       localStorage.setItem(SERVE_STATE_KEY, JSON.stringify(state.serveState));
     }
+    document.dispatchEvent(new CustomEvent('cookbook:state-synced', { detail: state }));
     return true;
   } catch { return false; }
 }
@@ -1332,17 +1443,11 @@ async function _openServeEditForTask(task, cmdOverride, fieldOverrides = null) {
   if (fieldOverrides && typeof fieldOverrides === 'object') {
     fields = { ...(fields || {}), ...fieldOverrides };
   }
-  // Switch the active server to the one this serve ran on (mirrors _openEdit).
+  fields = { ...(fields || {}), _replaceTaskId: task.sessionId };
-  const _tHost = task.remoteHost || '';
+  // Switch the active server to the exact profile this serve ran on. The
-  _envState.remoteHost = _tHost;
+  // dropdown stores stable srv: keys, not raw host strings, so preserving only
-  const _tSrv = _serverByVal(_envState.remoteServerKey || _tHost)
+  // task.remoteHost can relaunch against the local container by accident.
-    || _envState.servers.find(s => s.host === _tHost);
+  _selectTaskServer(task);
-  if (_tSrv) { _envState.env = _tSrv.env || 'none'; _envState.envPath = _tSrv.envPath || ''; _envState.platform = _tSrv.platform || ''; }
-  else if (!_tHost) { _envState.env = 'none'; _envState.envPath = ''; _envState.platform = ''; }
-  document.querySelectorAll('#hwfit-server-select, #hwfit-dl-server, #hwfit-cache-server, #hwfit-deps-server').forEach(sel => {
-    if (!sel || sel.tagName !== 'SELECT') return;
-    sel.value = _tHost || 'local';
-  });
   try {
     const { openServePanelForRepo } = await import('./cookbookServe.js');
     await openServePanelForRepo(repo, fields);
@@ -1553,6 +1658,20 @@ export async function _launchServeTask(shortName, repo, cmd, fields, hostOverrid
   const _serverMetaKey = _targetKey || (_hsrv && _serverKey ? _serverKey(_hsrv) : '') || (_host || 'local');
   const _serverMetaName = targetMeta?.serverName || _hsrv.name || (_host ? _host : 'Local');
   const _hplatform = _host ? (_hsrv.platform || '') : (_envState.platform || '');
+  const _replaceTaskId = fields?._replaceTaskId || '';
+  if (_replaceTaskId) {
+    try {
+      const _old = _loadTasks().find(t => t.sessionId === _replaceTaskId);
+      if (_old && _old.type === 'serve') {
+        await fetch('/api/shell/exec', {
+          method: 'POST', credentials: 'same-origin',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({ command: _tmuxGracefulKill(_old) }),
+        });
+        _removeTask(_old.sessionId);
+      }
+    } catch {}
+  }
 
   // Replace any serve already targeting this same host:port — you can't run two
   // servers on one port, so re-serving (or retrying) should stop & remove the
@@ -1750,7 +1869,7 @@ export function _renderRunningTab() {
       '<div style="display:flex;align-items:baseline;gap:8px;margin-bottom:2px;">' +
       '<h2 style="margin:0;padding:0;line-height:1;">Active <span id="running-count" class="memory-count" style="font-size:0.6em;opacity:0.6;font-weight:normal">' + activeCount + '</span></h2>' +
       '</div>' +
-      '<p class="memory-desc doclib-desc" style="margin-top:6px;">Active downloads and serving processes.</p>' +
+      '<p class="memory-desc doclib-desc" style="margin-top:6px;">Active downloads, installs and model launches.</p>' +
       '</div>';
     const firstGroup = body.querySelector('.cookbook-group');
     if (firstGroup) body.insertBefore(group, firstGroup);
@@ -1863,6 +1982,7 @@ export function _renderRunningTab() {
         return;
       }
       if (!await window.styledConfirm(`Clear ${toRemove.length} finished task${toRemove.length === 1 ? '' : 's'} on ${_serverName(host)}?`, { confirmText: 'Clear' })) return;
+      toRemove.forEach(t => _tombstoneTask(t.sessionId));
       const remaining = allTasks.filter(t => _taskServerKey(t) !== host || !_canClearTask(t));
       _saveTasks(remaining);
       // Fade/slide each finished card out (same exit as the per-card clear)
@@ -2000,11 +2120,12 @@ export function _renderRunningTab() {
 
     const _bdg = _taskBadge(task);
     const _bdgTitle = (task._unreachable && task.status === 'running') ? ' title="Server not responding — it may have crashed"' : '';
+    const displayName = _taskDisplayName(task);
     el.innerHTML = `
       <div class="cookbook-task-header">
         <span class="cookbook-task-type${(task.status === 'done' && task.type === 'download') ? ' cookbook-task-type-done' : ''}" data-type="${esc(task.type)}">${esc((task.status === 'done' && task.type === 'download') ? 'finished' : task.type)}</span>
-        <span class="cookbook-task-name">${modelLogo(task.name)}${esc(task.name)}</span>
+        <span class="cookbook-task-name">${modelLogo(task.name)}${esc(displayName)}</span>
-        <span class="cookbook-task-indicator"><span class="cookbook-task-wave" style="display:${task.status === 'running' ? '' : 'none'}"></span><span class="cookbook-task-check" title="Clear" style="display:${_canClearTask(task) ? '' : 'none'}"><svg class="cookbook-task-check-ico" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="#50fa7b" stroke-width="3" stroke-linecap="round" stroke-linejoin="round"><polyline points="20 6 9 17 4 12"/></svg><svg class="cookbook-task-clear-ico" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="3" stroke-linecap="round" stroke-linejoin="round"><line x1="18" y1="6" x2="6" y2="18"/><line x1="6" y1="6" x2="18" y2="18"/></svg><span class="cookbook-task-done-label">${esc(_clearPillLabel(task))}</span><span class="cookbook-task-clear-label">clear</span></span></span>
+        <span class="cookbook-task-indicator"><span class="cookbook-task-wave" style="display:${task.status === 'running' ? '' : 'none'}"></span>${_canLaunchDownloadedTask(task) ? '<button type="button" class="cookbook-task-serve-btn" title="Open in Launch"><svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.4" stroke-linecap="round" stroke-linejoin="round"><polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"/></svg><span>Launch</span></button>' : ''}<span class="cookbook-task-check" title="Clear" style="display:${_canClearTask(task) ? '' : 'none'}"><svg class="cookbook-task-check-ico" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="#50fa7b" stroke-width="3" stroke-linecap="round" stroke-linejoin="round"><polyline points="20 6 9 17 4 12"/></svg><svg class="cookbook-task-clear-ico" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="3" stroke-linecap="round" stroke-linejoin="round"><line x1="18" y1="6" x2="6" y2="18"/><line x1="6" y1="6" x2="18" y2="18"/></svg><span class="cookbook-task-done-label">${esc(_clearPillLabel(task))}</span><span class="cookbook-task-clear-label">clear</span></span></span>
         <button type="button" class="cookbook-task-start-now" title="Start this queued download now" style="display:${(task.type === 'download' && task.status === 'queued') ? '' : 'none'}"><svg width="11" height="11" viewBox="0 0 24 24" fill="currentColor" aria-hidden="true"><polygon points="8 5 19 12 8 19 8 5"/></svg><span>start now</span></button>
         <span class="cookbook-task-status ${_bdg.cls}"${_bdgTitle}>${esc(_bdg.text)}</span>
         <button class="cookbook-task-menu-btn" title="Actions">&#8942;</button>
@@ -2076,19 +2197,11 @@ export function _renderRunningTab() {
           e.stopPropagation();
           const repo = task.payload?.repo_id || task.name;
           if (!repo) { uiModule.showToast('No model info on this task'); return; }
-          // Point the active server at the one it downloaded to.
+          // Point the active server at the exact profile it downloaded to.
-          const _tHost = task.remoteHost || '';
+          _selectTaskServer(task);
-          _envState.remoteHost = _tHost;
-          const _tSrv = _serverByVal(_envState.remoteServerKey || _tHost)
-            || _envState.servers.find(s => s.host === _tHost);
-          if (_tSrv) { _envState.env = _tSrv.env || 'none'; _envState.envPath = _tSrv.envPath || ''; _envState.platform = _tSrv.platform || ''; }
-          else if (!_tHost) { _envState.env = 'none'; _envState.envPath = ''; _envState.platform = ''; }
-          document.querySelectorAll('#hwfit-server-select, #hwfit-dl-server, #hwfit-cache-server, #hwfit-deps-server').forEach(sel => {
-            if (sel && sel.tagName === 'SELECT') sel.value = _tHost || 'local';
-          });
           try {
             const { openServePanelForRepo } = await import('./cookbookServe.js');
-            await openServePanelForRepo(repo);
+            await openServePanelForRepo(repo, _downloadServeFields(task));
             // Serving it supersedes the finished download — clear the card from
             // the Running tab (smooth exit) now that we've jumped to Serve.
             _animateOutThenRemove(el, task.sessionId);
@@ -3558,7 +3671,9 @@ async function _probeEndpointUntilOnline(epId, host, port) {
     try {
       // Hit the probe endpoint — it re-probes server-side and updates
       // cached_models. We consume (and discard) the SSE stream.
-      await fetch(`/api/model-endpoints/${epId}/probe`, { credentials: 'same-origin' }).then(r => r.text()).catch(() => {});
+      const probeRes = await fetch(`/api/model-endpoints/${epId}/probe`, { credentials: 'same-origin' }).catch(() => null);
+      if (probeRes && probeRes.status === 404) return;
+      if (probeRes) await probeRes.text().catch(() => {});
       const eps = await fetch('/api/model-endpoints', { credentials: 'same-origin' }).then(r => r.json()).catch(() => []);
       const ep = (eps || []).find(e => e.id === epId);
       if (ep && (ep.models || []).length) {

static/js/cookbookServe.js

@@ -477,7 +477,9 @@ function _estimateLlamaContextFit(model, fields, modelCtxMax, modelWeightsGb = 0
 }
 
 function _selectedServeTarget(panel) {
-  const select = document.getElementById('hwfit-server-select') || document.getElementById('hwfit-dl-server');
+  const select = panel?.querySelector?.('#hwfit-server-select')
+    || document.getElementById('hwfit-server-select')
+    || document.getElementById('hwfit-dl-server');
   const servers = Array.isArray(_envState.servers) ? _envState.servers : [];
   let host = _envState.remoteHost || '';
   let server = host ? (_serverByVal?.(_envState.remoteServerKey || host) || servers.find(s => s.host === host)) : null;
@@ -643,6 +645,122 @@ function _ggufFileLabel(file) {
   return `${quant}${base}${size || split ? ` (${[size, split.replace(/^, /, '')].filter(Boolean).join(', ')})` : ''}${role}`;
 }
 
+function _ggufTaskDisplayPart(model, relPath) {
+  const rel = String(relPath || '');
+  if (!rel) return '';
+  const file = _ggufFilesForModel(model).find(f => f.rel_path === rel);
+  if (file?.quant) return String(file.quant).toUpperCase().replace(/^UD-/, '');
+  const parts = rel.split('/').filter(Boolean);
+  const base = parts[parts.length - 1] || '';
+  const parent = parts.length > 1 ? parts[parts.length - 2] : '';
+  const text = `${parent} ${base}`;
+  const quant = text.match(/\b(?:UD-)?(?:IQ[1-8]_[A-Z0-9]+|Q[2-8]_K_[MLS]|Q[2-8]_[0-9A-Z]+|Q[2-8])\b/i);
+  if (quant) return quant[0].toUpperCase().replace(/^UD-/, '');
+  return base.replace(/\.gguf$/i, '').replace(/-\d{5}-of-\d{5}$/i, '');
+}
+
+function _serveTaskDisplayName(shortName, model, fields) {
+  const name = String(shortName || '').trim();
+  const backend = String(fields?.backend || '').toLowerCase();
+  if (backend !== 'llamacpp' && backend !== 'ollama') return name;
+  const part = _ggufTaskDisplayPart(model, fields?.gguf_file);
+  return part && !name.includes(` · ${part}`) ? `${name} · ${part}` : name;
+}
+
+function _safeGgufRelPath(relPath) {
+  const rel = String(relPath || '').replace(/\\/g, '/').replace(/^\/+/, '');
+  if (!rel || rel.startsWith('../') || rel.includes('/../') || rel === '..') return '';
+  if (rel.includes('\0')) return '';
+  return rel;
+}
+
+function _ggufDeleteChoice(repo, files) {
+  return new Promise(resolve => {
+    let overlay = document.getElementById('cookbook-gguf-delete-overlay');
+    if (!overlay) {
+      overlay = document.createElement('div');
+      overlay.id = 'cookbook-gguf-delete-overlay';
+      overlay.className = 'modal hidden';
+      overlay.innerHTML =
+        '<div class="modal-content styled-confirm-box cookbook-gguf-delete-box" role="dialog" aria-modal="true" aria-labelledby="cookbook-gguf-delete-title">' +
+          '<div class="modal-header"><h4 id="cookbook-gguf-delete-title">Delete GGUF files</h4></div>' +
+          '<div class="modal-body">' +
+            '<p id="cookbook-gguf-delete-msg"></p>' +
+            '<div id="cookbook-gguf-delete-list" class="cookbook-gguf-delete-list"></div>' +
+          '</div>' +
+          '<div class="modal-footer cookbook-gguf-delete-actions">' +
+            '<button type="button" id="cookbook-gguf-delete-cancel" class="confirm-btn confirm-btn-secondary">Cancel</button>' +
+            '<button type="button" id="cookbook-gguf-delete-repo" class="confirm-btn confirm-btn-secondary">Whole repo</button>' +
+            '<button type="button" id="cookbook-gguf-delete-selected" class="confirm-btn confirm-btn-danger">Delete selected</button>' +
+          '</div>' +
+        '</div>';
+      document.body.appendChild(overlay);
+    }
+
+    const safeFiles = files
+      .map(f => ({ ...f, rel_path: _safeGgufRelPath(f.rel_path) }))
+      .filter(f => f.rel_path);
+    const msg = overlay.querySelector('#cookbook-gguf-delete-msg');
+    const list = overlay.querySelector('#cookbook-gguf-delete-list');
+    const cancelBtn = overlay.querySelector('#cookbook-gguf-delete-cancel');
+    const repoBtn = overlay.querySelector('#cookbook-gguf-delete-repo');
+    const selectedBtn = overlay.querySelector('#cookbook-gguf-delete-selected');
+    const prevFocus = document.activeElement;
+
+    msg.textContent = `${repo} has multiple GGUF files. Pick what to delete.`;
+    list.innerHTML = safeFiles.map((file, idx) => {
+      const label = esc ? esc(_ggufFileLabel(file)) : _ggufFileLabel(file);
+      const rel = esc ? esc(file.rel_path) : file.rel_path;
+      return `<label class="cookbook-gguf-delete-row">
+        <input class="cookbook-gguf-delete-cb" type="checkbox" value="${idx}">
+        <span class="cookbook-gguf-delete-main">${label}</span>
+        <span class="cookbook-gguf-delete-path">${rel}</span>
+      </label>`;
+    }).join('');
+
+    function cleanup(result) {
+      overlay.classList.add('hidden');
+      overlay.style.display = 'none';
+      cancelBtn.removeEventListener('click', onCancel);
+      repoBtn.removeEventListener('click', onRepo);
+      selectedBtn.removeEventListener('click', onSelected);
+      overlay.removeEventListener('click', onBackdrop);
+      document.removeEventListener('keydown', onKey);
+      try { prevFocus && prevFocus.focus && prevFocus.focus(); } catch {}
+      resolve(result);
+    }
+    function onCancel() { cleanup(null); }
+    function onRepo() { cleanup({ mode: 'repo' }); }
+    function onSelected() {
+      const selected = [...list.querySelectorAll('input[type="checkbox"]:checked')]
+        .map(input => safeFiles[Number(input.value)])
+        .filter(Boolean);
+      if (!selected.length) {
+        uiModule.showToast?.('Select at least one GGUF file.');
+        return;
+      }
+      cleanup({ mode: 'files', files: selected });
+    }
+    function onBackdrop(e) { if (e.target === overlay) cleanup(null); }
+    function onKey(e) {
+      if (e.key === 'Escape') {
+        e.preventDefault();
+        e.stopPropagation();
+        cleanup(null);
+      }
+    }
+
+    cancelBtn.addEventListener('click', onCancel);
+    repoBtn.addEventListener('click', onRepo);
+    selectedBtn.addEventListener('click', onSelected);
+    overlay.addEventListener('click', onBackdrop);
+    document.addEventListener('keydown', onKey);
+    overlay.classList.remove('hidden');
+    overlay.style.display = '';
+    selectedBtn.focus();
+  });
+}
+
 function _shellPathExpr(path) {
   const s = String(path || '');
   if (s === '~') return '${HOME}';
@@ -729,7 +847,7 @@ function _rerenderCachedModels() {
       ? ` <span class="cookbook-serve-downloading-pill${_isDlActive ? '' : ' is-stalled'}" title="${_isDlActive ? 'Download in progress' : 'Download stalled — retry to resume'}">${_isDlActive ? 'downloading' : 'stalled'}</span>`
       : '';
     const _favoritePill = _isFavorite ? ' <span class="memory-cat-badge memory-cat-pinned cookbook-serve-fav-badge">pinned</span>' : '';
-    html += `<div class="memory-item-title"${_mc ? ` style="color:${_mc}"` : ''}>${modelLogo(m.repo_id)}${esc(shortName)}${_favoritePill}${hfLink ? ` <a href="${esc(hfLink)}" target="_blank" rel="noopener" class="cookbook-hf-link">HF ↗</a>` : ''}${_runningPill}${_downloadingPill}</div>`;
+    html += `<div class="memory-item-title cookbook-serve-title"${_mc ? ` style="color:${_mc}"` : ''}><span class="cookbook-serve-title-name">${modelLogo(m.repo_id)}${esc(shortName)}</span>${_favoritePill}${hfLink ? ` <a href="${esc(hfLink)}" target="_blank" rel="noopener" class="cookbook-hf-link">HF ↗</a>` : ''}${_runningPill}${_downloadingPill}</div>`;
     html += `<div class="memory-item-meta" style="font-size:10px;opacity:0.4;margin-top:2px;">${metaParts.join(' \u00b7 ')}</div>`;
     html += `</div>`;
     const _bk = _detectBackend(m).backend;
@@ -962,6 +1080,11 @@ function _rerenderCachedModels() {
       const _isMiniMaxM3 = _isMiniMaxM3Model({ ...m, repo_id: repo });
       const _isMiniMaxM2 = _isMiniMaxM2Model({ ...m, repo_id: repo });
       const _isMiniMaxMSeries = _isMiniMaxM3 || _isMiniMaxM2;
+      const _toolParserDefault = _detectToolParser(repo);
+      const _isStepFunStep = _toolParserDefault === 'step3p5';
+      const _nativeToolDefault = _isMiniMaxMSeries || _isStepFunStep;
+      const _reasoningDefault = _isMiniMaxMSeries || _isStepFunStep;
+      const _expertParallelDefault = _isMiniMaxMSeries || _isStepFunStep;
       const svm = (k, def) => (_modelSs && _hasOwn(_modelSs, k)) ? _modelSs[k] : def;
       const _serveTarget = _selectedServeTarget();
       const _backendChoices = _backendChoicesForTarget(_serveTarget);
@@ -993,8 +1116,15 @@ function _rerenderCachedModels() {
       const _l = (name, tip) => `<span>${name}<span class="hwfit-hint" title="${tip}">?</span></span>`;
       const _ggufChoices = _runnableGgufFiles(m);
       const _savedGguf = String(sv('gguf_file', '') || '');
+      const _preferredGgufInclude = String(sv('_preferredGgufInclude', '') || '').replace(/\*/g, '').toLowerCase();
+      const _preferredGguf = _preferredGgufInclude
+        ? (_ggufChoices.find(f => String(f.rel_path || '').toLowerCase().includes(_preferredGgufInclude))
+          || _ggufChoices.find(f => String(f.name || '').toLowerCase().includes(_preferredGgufInclude)))
+        : null;
       const _defaultGguf = _ggufChoices.some(f => f.rel_path === _savedGguf)
         ? _savedGguf
+        : (_preferredGguf?.rel_path || '')
+          ? _preferredGguf.rel_path
         : (_ggufChoices[0]?.rel_path || '');
       const _ggufOptions = _ggufChoices.map(f =>
         `<option value="${esc(f.rel_path)}"${f.rel_path === _defaultGguf ? ' selected' : ''}>${esc(_ggufFileLabel(f))}</option>`
@@ -1026,6 +1156,10 @@ function _rerenderCachedModels() {
         + `</div>`;
 
       let panelHtml = `<div class="hwfit-serve-panel">`;
+      const _replaceTaskId = String(sv('_replaceTaskId', '') || '');
+      if (_replaceTaskId) {
+        panelHtml += `<input type="hidden" class="hwfit-sf" data-field="_replaceTaskId" value="${esc(_replaceTaskId)}" />`;
+      }
       // Runtime-readiness note pinned at the top of the serve area so the
       // user sees "vLLM ready on …" before scrolling into the configure
       // form. Hidden until the readiness probe returns. The × button
@@ -1202,20 +1336,20 @@ function _rerenderCachedModels() {
       const _rp_name = _rp_flag ? _rp_flag.split(' ')[1] : '';
       panelHtml += `<div class="hwfit-serve-checks hwfit-backend-vllm hwfit-backend-sglang">`;
       panelHtml += `<label class="hwfit-sf-cb"><input type="checkbox" class="hwfit-sf" data-field="trust_remote"${sv('trust_remote',_isMiniMaxMSeries)?' checked':''} /> Trust Remote Code${_h('Allow model to run custom code from HuggingFace')}</label>`;
-      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm"><input type="checkbox" class="hwfit-sf" data-field="auto_tool"${sv('auto_tool',_isMiniMaxMSeries)?' checked':''} /> Auto Tool Choice${_h('Enable function/tool calling for agent mode')}</label>`;
+      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm hwfit-backend-sglang"><input type="checkbox" class="hwfit-sf" data-field="auto_tool"${sv('auto_tool',_nativeToolDefault)?' checked':''} /> Auto Tool Choice${_h('Enable function/tool calling for agent mode')}</label>`;
       // Always-render the Reasoning Parser, Expert Parallel, and MoE Env
       // checkboxes — the model-family detection above is a hint, not a
       // hard gate. User asked to keep these visible regardless so that
       // a borderline-undetected MoE/reasoning model can still toggle
       // them without dropping back to the raw command box.
-      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm"><input type="checkbox" class="hwfit-sf" data-field="reasoning_parser" data-parser="${_rp_name || ''}"${sv('reasoning_parser',_isMiniMaxMSeries)?' checked':''} /> Reasoning Parser${_rp_name ? ` <span class="hwfit-parser-tag">${_rp_name}</span>` : ''}${_h('Splits <think> tokens into a separate channel. The tag (when shown) is the auto-detected parser; edit the command if you need a different one.')}</label>`;
+      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm hwfit-backend-sglang"><input type="checkbox" class="hwfit-sf" data-field="reasoning_parser" data-parser="${_rp_name || ''}"${sv('reasoning_parser',_reasoningDefault)?' checked':''} /> Reasoning Parser${_rp_name ? ` <span class="hwfit-parser-tag">${_rp_name}</span>` : ''}${_h('Splits <think> tokens into a separate channel. The tag (when shown) is the auto-detected parser; edit the command if you need a different one.')}</label>`;
       panelHtml += `<label class="hwfit-sf-cb"><input type="checkbox" class="hwfit-sf" data-field="enforce_eager"${sv('enforce_eager',false)?' checked':''} /> Enforce Eager${_h('Disable CUDA graphs. Slower but uses less memory')}</label>`;
       panelHtml += `<label class="hwfit-sf-cb"><input type="checkbox" class="hwfit-sf" data-field="prefix_cache"${sv('prefix_cache',false)?' checked':''} /> Prefix Caching${_h('Cache shared prompt prefixes across requests')}</label>`;
       // Inline the previously-second vLLM checks row so Expert Parallel /
       // Speculative / MoE Env sit next to Prefix Caching with no gap. All
       // three are vLLM-only — class-gated so they hide on SGLang. Always
       // render so the user can flip them on for any MoE model.
-      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm"><input type="checkbox" class="hwfit-sf" data-field="expert_parallel"${sv('expert_parallel',_isMiniMaxMSeries)?' checked':''} /> Expert Parallel${_h('MoE: shard expert layers across GPUs. Helps for MiniMax M-series, Qwen3 A3B/A10B/A22B MoE, DeepSeek V3+/R1. Ignored / wasteful on dense models.')}</label>`;
+      panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm hwfit-backend-sglang"><input type="checkbox" class="hwfit-sf" data-field="expert_parallel"${sv('expert_parallel',_expertParallelDefault)?' checked':''} /> Expert Parallel${_h('MoE: shard expert layers across GPUs. Helps for MiniMax M-series, StepFun Step-3, Qwen3 A3B/A10B/A22B MoE, DeepSeek V3+/R1. Ignored / wasteful on dense models.')}</label>`;
       panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm"><input type="checkbox" class="hwfit-sf" data-field="language_model_only"${sv('language_model_only',_isMiniMaxM3)?' checked':''} /> Language Model Only${_h('vLLM --language-model-only. Needed by MiniMax M3 text serving when the repo also contains VL components.')}</label>`;
       panelHtml += `<label class="hwfit-sf-cb hwfit-backend-vllm"><input type="checkbox" class="hwfit-sf" data-field="disable_custom_all_reduce"${sv('disable_custom_all_reduce',_isMiniMaxM3)?' checked':''} /> Disable Custom All Reduce${_h('vLLM --disable-custom-all-reduce. Useful for some 8-GPU/nightly configurations.')}</label>`;
       {
@@ -2870,11 +3004,11 @@ function _rerenderCachedModels() {
             // preflight and let the launch silently fall to CPU.
             let _hwGpus = [];
             try {
-              const _gh = (_selectedServeTarget.host || '').trim();
+              const _gh = (launchTarget.host || '').trim();
               const _gp = new URLSearchParams();
               if (_gh) {
                 _gp.set('host', _gh);
-                const _sp = (_serverByVal?.(_selectedServeTarget.serverKey || _gh) || {}).port;
+                const _sp = (_serverByVal?.(launchTarget.serverKey || _gh) || {}).port;
                 if (_sp) _gp.set('ssh_port', _sp);
               }
               const _gr = await fetch('/api/cookbook/gpus' + (_gp.toString() ? '?' + _gp : ''), { credentials: 'same-origin' });
@@ -3069,6 +3203,7 @@ function _rerenderCachedModels() {
           try { cur = JSON.parse(localStorage.getItem(SERVE_STATE_KEY)) || {}; } catch {}
           const byRepo = (cur && cur._byRepo && typeof cur._byRepo === 'object') ? cur._byRepo : {};
           const _saved = { ...serveState, _forceBackend: true };
+          delete _saved._replaceTaskId;
           byRepo[repo] = _saved;
           localStorage.setItem(SERVE_STATE_KEY, JSON.stringify({ _byRepo: byRepo, _lastUsed: _saved }));
         } catch {}
@@ -3127,7 +3262,8 @@ function _rerenderCachedModels() {
           await _withSpinner(_launchBtn, async () => {
             // Pass the exact form values so the running task can be re-opened
             // in the Serve panel pre-filled with these settings (Edit button).
-            await _launchServeTask(shortName, repo, launchCmd, serveState, serveHost, { serverKey: serveServerKey, serverName: serveServerName });
+            const taskDisplayName = _serveTaskDisplayName(shortName, m, serveState);
+            await _launchServeTask(taskDisplayName, repo, launchCmd, serveState, serveHost, { serverKey: serveServerKey, serverName: serveServerName });
           });
         } finally {
           _envState.env = origEnv;
@@ -3188,7 +3324,6 @@ function _resolveCacheHost() {
 }
 
 async function _deleteCachedModel(repo, itemEl, skipConfirm = false, model = null) {
-  if (!skipConfirm && !(await uiModule.styledConfirm(`Delete ${repo} from cache?`, { confirmText: 'Delete', danger: true }))) return;
   const m = model || _cachedAllModels.find(x => x.repo_id === repo);
   // Delete the EXACT on-disk path the scan reported. Models in a custom
   // model dir live at <path>/<repo>; HF-cache models at
@@ -3204,13 +3339,32 @@ async function _deleteCachedModel(repo, itemEl, skipConfirm = false, model = nul
   } else {
     target = `~/.cache/huggingface/hub/models--${repo.replace(/\//g, '--')}`;
   }
+  let deleteChoice = { mode: 'repo' };
+  const ggufFiles = _ggufFilesForModel(m);
+  if (!skipConfirm) {
+    if (ggufFiles.length > 1) {
+      deleteChoice = await _ggufDeleteChoice(repo, ggufFiles);
+      if (!deleteChoice) return;
+    } else if (!(await uiModule.styledConfirm(`Delete ${repo} from cache?`, { confirmText: 'Delete', danger: true }))) {
+      return;
+    }
+  }
   const host = _resolveCacheHost();
   let cmd;
   if (_isWindows()) {
     const winTarget = target.startsWith('~')
       ? target.replace(/^~/, '$env:USERPROFILE').replace(/\//g, '\\')
       : target.replace(/\//g, '\\');
-    cmd = `Remove-Item -Recurse -Force "${winTarget}" -ErrorAction SilentlyContinue`;
+    if (deleteChoice.mode === 'files') {
+      const targets = deleteChoice.files
+        .map(f => _safeGgufRelPath(f.rel_path))
+        .filter(Boolean)
+        .map(rel => `${winTarget}\\${rel.replace(/\//g, '\\')}`);
+      if (!targets.length) return;
+      cmd = targets.map(p => `Remove-Item -Force "${p.replace(/"/g, '\\"')}" -ErrorAction SilentlyContinue`).join('; ');
+    } else {
+      cmd = `Remove-Item -Recurse -Force "${winTarget}" -ErrorAction SilentlyContinue`;
+    }
     if (host) {
       const pf = _sshPrefix(_getPort(host));
       cmd = `ssh ${pf}${host} "powershell -Command \\"${cmd}\\""`;
@@ -3219,7 +3373,16 @@ async function _deleteCachedModel(repo, itemEl, skipConfirm = false, model = nul
     // $HOME expands inside double quotes; ~ would not, so normalize the
     // fallback. Quoting also handles spaces in custom model-dir paths.
     const unixTarget = target.startsWith('~') ? target.replace(/^~/, '$HOME') : target;
-    cmd = `rm -rf "${unixTarget}"`;
+    if (deleteChoice.mode === 'files') {
+      const targets = deleteChoice.files
+        .map(f => _safeGgufRelPath(f.rel_path))
+        .filter(Boolean)
+        .map(rel => `${target.replace(/\/+$/, '')}/${rel}`);
+      if (!targets.length) return;
+      cmd = `rm -f ${targets.map(p => _shellPathExpr(p)).join(' ')} && find ${_shellPathExpr(target)} -type d -empty -delete`;
+    } else {
+      cmd = `rm -rf "${unixTarget}"`;
+    }
     if (host) cmd = _sshCmd(host, cmd, _getPort(host));
   }
   // Deleting a large model (tens/hundreds of GB) can take a while, especially
@@ -3244,7 +3407,13 @@ async function _deleteCachedModel(repo, itemEl, skipConfirm = false, model = nul
       body: JSON.stringify({ command: cmd }),
     });
     if (!res.ok) { uiModule.showError(`Delete failed (${res.status})`); return; }
-    if (itemEl) {
+    if (deleteChoice.mode === 'files') {
+      if (m && Array.isArray(m.gguf_files)) {
+        const removed = new Set(deleteChoice.files.map(f => _safeGgufRelPath(f.rel_path)));
+        m.gguf_files = m.gguf_files.filter(f => !removed.has(_safeGgufRelPath(f.rel_path)));
+      }
+      await _fetchCachedModels(false);
+    } else if (itemEl) {
       itemEl.querySelector('.cookbook-delete-overlay')?.remove();
       itemEl.style.transition = 'opacity 0.24s ease, transform 0.24s ease, max-height 0.28s ease, padding 0.28s ease, margin 0.28s ease';
       itemEl.style.maxHeight = `${Math.max(itemEl.getBoundingClientRect().height, itemEl.scrollHeight)}px`;
@@ -3258,9 +3427,9 @@ async function _deleteCachedModel(repo, itemEl, skipConfirm = false, model = nul
       requestAnimationFrame(() => { itemEl.style.maxHeight = '0'; });
       await new Promise(resolve => setTimeout(resolve, 300));
       if (itemEl.parentElement) itemEl.remove();
+      // Drop from the in-memory list so a re-render/filter doesn't resurrect it.
+      _cachedAllModels = _cachedAllModels.filter(x => x.repo_id !== repo);
     }
-    // Drop from the in-memory list so a re-render/filter doesn't resurrect it.
-    _cachedAllModels = _cachedAllModels.filter(x => x.repo_id !== repo);
   } catch (e) {
     uiModule.showError('Delete failed: ' + (e && e.message ? e.message : e));
   } finally {

static/js/modelPicker.js

@@ -77,6 +77,7 @@ function _handlePickerKeydown(e, listEl, itemSelector, closeFn) {
 // Dependencies injected via initModelPicker()
 let _deps = null;
 let _autoSelectingDefault = false;
+let _defaultChatPickInFlight = false;
 
 function _modelExists(modelId, url) {
   if (!modelId || !window.modelsModule || !window.modelsModule.getCachedItems) return false;
@@ -91,6 +92,43 @@ function _modelExists(modelId, url) {
   });
 }
 
+async function _ensureDefaultPendingChat() {
+  if (!_deps || _defaultChatPickInFlight) return;
+  if (_deps.getCurrentSessionId && _deps.getCurrentSessionId()) return;
+  const pending = _deps.getPendingChat && _deps.getPendingChat();
+  if (pending && pending.modelId) return;
+  _defaultChatPickInFlight = true;
+  try {
+    let dc = null;
+    try {
+      const res = await fetch(`${API_BASE}/api/default-chat`, { credentials: 'same-origin' });
+      if (res.ok) dc = await res.json();
+    } catch (_) {}
+    if (dc && dc.endpoint_url && dc.model) {
+      _deps.setPendingChat({
+        url: dc.endpoint_url,
+        modelId: dc.model,
+        endpointId: dc.endpoint_id || '',
+      });
+      try { window.__odysseusDefaultChat = dc; } catch (_) {}
+      updateModelPicker();
+      return;
+    }
+    // No configured default: preserve the old convenience fallback.
+    if (window.modelsModule && window.modelsModule.getCachedItems) {
+      const items = window.modelsModule.getCachedItems();
+      const first = items.find(item => !item.offline && ((item.models || []).length || (item.models_extra || []).length));
+      if (first) {
+        const models = (first.models || []).concat(first.models_extra || []);
+        _deps.setPendingChat({ url: first.url, modelId: models[0], endpointId: first.endpoint_id });
+        updateModelPicker();
+      }
+    }
+  } finally {
+    _defaultChatPickInFlight = false;
+  }
+}
+
 /**
  * Initialize the model picker dropdown.
  * @param {Object} deps
@@ -710,25 +748,7 @@ export function updateModelPicker() {
     }
   }
   if (!modelId && !_autoSelectingDefault && window.modelsModule && window.modelsModule.getCachedItems) {
-    const items = window.modelsModule.getCachedItems();
+    _ensureDefaultPendingChat();
-    const first = items.find(item => !item.offline && ((item.models || []).length || (item.models_extra || []).length));
-    if (first) {
-      const models = (first.models || []).concat(first.models_extra || []);
-      modelId = models[0];
-      if (!currentSessionId) {
-        _deps.setPendingChat({ url: first.url, modelId, endpointId: first.endpoint_id });
-      } else {
-        if (s) { s.model = modelId; s.endpoint_url = first.url; }
-        _autoSelectingDefault = true;
-        const fd = new FormData();
-        fd.append('model', modelId);
-        fd.append('endpoint_url', first.url || '');
-        if (first.endpoint_id) fd.append('endpoint_id', first.endpoint_id);
-        fetch(`${API_BASE}/api/session/${currentSessionId}`, { method: 'PATCH', body: fd })
-          .catch(() => {})
-          .finally(() => { _autoSelectingDefault = false; });
-      }
-    }
   }
 
   const displayName = modelId ? modelId.split('/').pop() : 'Select model';

static/js/notes.js

@@ -1896,10 +1896,6 @@ function _renderNotes() {
       ${_hasItems(note) ? `<div class="note-cl-quickadd"><input type="text" class="note-cl-quickadd-input" placeholder="+ Add item" data-note-id="${note.id}" /></div>` : ''}
       ${reminderTagHtml}
       ${noteTags.length ? `<div class="note-card-label">${noteTags.map(t => `<button type="button" class="note-card-label-chip" data-note-label-filter="${_esc(t)}" title="Filter #${_esc(t)}">#${_esc(t)}</button>`).join(' ')}</div>` : ''}
-      ${note.agent_session_id ? `<button class="note-agent-tag" data-note-id="${note.id}" data-session-id="${_esc(note.agent_session_id)}" title="Open the agent's chat for this note">
-        <svg width="11" height="11" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 8V4H8"/><rect x="4" y="8" width="16" height="12" rx="2"/><path d="M2 14h2M20 14h2M15 13v2M9 13v2"/></svg>
-        <span>Agent</span>
-      </button>` : ''}
       <div class="note-card-actions">
         <div class="note-card-colors">${colorDots}</div>
         <span style="flex:1"></span>
@@ -2304,16 +2300,6 @@ function _bindCardEvents(body) {
       _openNoteCornerMenu(btn);
     });
   });
-  // Agent tag — opens the chat session the agent ran for this note.
-  body.querySelectorAll('.note-agent-tag').forEach(tag => {
-    tag.addEventListener('click', (e) => {
-      e.preventDefault();
-      e.stopPropagation();
-      const sid = tag.dataset.sessionId;
-      const _sm = window.sessionModule;
-      if (sid && _sm && _sm.selectSession) { closePanel(); _sm.selectSession(sid); }
-    });
-  });
   body.querySelectorAll('.note-card-label-chip').forEach(chip => {
     chip.addEventListener('click', (e) => {
       e.preventDefault();
@@ -4383,18 +4369,16 @@ function _openTodoAgentMenu(btn) {
   const noteId = btn.dataset.noteId;
   const idx = parseInt(btn.dataset.idx);
   const sid = btn.dataset.sessionId || '';
-  const title = btn.dataset.agentTitle || 'Agent chat';
   const menu = document.createElement('div');
   menu.className = 'note-corner-menu-dropdown note-agent-item-menu';
   menu.innerHTML = `
-    <div class="ncm-title">${_esc(title)}</div>
     ${sid ? `<button type="button" class="ncm-item" data-act="open">
       <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M15 3h6v6"/><path d="M10 14L21 3"/><path d="M18 13v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/></svg>
-      <span>Open this agent chat</span>
+      <span>Open</span>
     </button>` : ''}
     <button type="button" class="ncm-item" data-act="run">
       <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 8V4H8"/><rect x="4" y="8" width="16" height="12" rx="2"/><path d="M2 14h2M20 14h2M15 13v2M9 13v2"/></svg>
-      <span>${sid ? 'Run again for this todo' : 'Start agent for this todo'}</span>
+      <span>${sid ? 'Run again' : 'Run Agent'}</span>
     </button>`;
   _positionNoteMenu(menu, btn);
   const openBtn = menu.querySelector('[data-act="open"]');

static/style.css

@@ -5324,6 +5324,84 @@ body.bg-pattern-sparkles {
     .confirm-btn-primary:hover { filter:brightness(1.15); }
     .confirm-btn-danger { background:var(--color-danger); color:#fff; border-color:transparent; }
     .confirm-btn-danger:hover { background:var(--color-error); }
+    #cookbook-gguf-delete-overlay {
+      background:rgba(0,0,0,0.5);
+      backdrop-filter:blur(4px);
+      pointer-events:auto !important;
+      z-index:99999 !important;
+      position:fixed !important;
+      inset:0 !important;
+    }
+    .cookbook-gguf-delete-box {
+      width:560px;
+      max-width:92vw;
+    }
+    .cookbook-gguf-delete-list {
+      display:flex;
+      flex-direction:column;
+      gap:6px;
+      max-height:42vh;
+      overflow:auto;
+      padding:2px 2px 4px;
+    }
+    .cookbook-gguf-delete-row {
+      display:grid;
+      grid-template-columns:18px minmax(0,1fr);
+      gap:7px 8px;
+      align-items:start;
+      padding:7px 8px;
+      border:1px solid var(--border);
+      border-radius:7px;
+      background:color-mix(in srgb, var(--panel, var(--bg)) 92%, var(--fg) 8%);
+      cursor:pointer;
+    }
+    .cookbook-gguf-delete-row:hover {
+      border-color:color-mix(in srgb, var(--accent-primary, var(--fg)) 45%, var(--border));
+    }
+    .cookbook-gguf-delete-cb {
+      -webkit-appearance:none;
+      appearance:none;
+      width:8px !important;
+      height:8px !important;
+      min-width:8px;
+      min-height:8px;
+      padding:0;
+      margin:4px 0 0;
+      border:1px solid var(--border);
+      border-radius:50%;
+      background:transparent;
+      box-sizing:content-box;
+      cursor:pointer;
+      transition:background 0.15s, border-color 0.15s, transform 0.12s;
+    }
+    .cookbook-gguf-delete-cb:hover {
+      border-color:var(--accent, var(--red));
+      transform:scale(1.12);
+    }
+    .cookbook-gguf-delete-cb:checked {
+      background:var(--accent, var(--red));
+      border-color:var(--accent, var(--red));
+    }
+    .cookbook-gguf-delete-main,
+    .cookbook-gguf-delete-path {
+      min-width:0;
+      overflow:hidden;
+      text-overflow:ellipsis;
+      white-space:nowrap;
+    }
+    .cookbook-gguf-delete-main {
+      font-size:0.86rem;
+      color:var(--fg);
+    }
+    .cookbook-gguf-delete-path {
+      grid-column:2;
+      margin-top:-2px;
+      font-size:0.74rem;
+      opacity:0.58;
+    }
+    .cookbook-gguf-delete-actions {
+      flex-wrap:wrap;
+    }
     /* Styled prompt — text-input dialog (used in place of window.prompt) */
     #styled-prompt-overlay {
       background:rgba(0,0,0,0.5);
@@ -19222,6 +19300,18 @@ body.gallery-selecting .gallery-dl-btn,
   background: color-mix(in srgb, var(--red) 20%, transparent);
 }
 .cookbook-gpu-kill:disabled { opacity: 0.4; cursor: wait; }
+.cookbook-serve-title {
+  display: flex;
+  align-items: center;
+  gap: 4px;
+  min-width: 0;
+}
+.cookbook-serve-title-name {
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
 .cookbook-hf-link {
   font-size: 9px;
   text-decoration: none;
@@ -19234,6 +19324,7 @@ body.gallery-selecting .gallery-dl-btn,
   vertical-align: 1px;
   letter-spacing: 0.3px;
   font-weight: 600;
+  flex-shrink: 0;
 }
 .cookbook-hf-link:hover {
   opacity: 0.8;
@@ -19626,6 +19717,9 @@ body.gallery-selecting .gallery-dl-btn,
   position: relative;
   top: -2px;
 }
+.cookbook-dep-reinstall {
+  top: -3px;
+}
 .cookbook-dep-rebuild:hover {
   background: color-mix(in srgb, var(--accent, var(--red)) 18%, transparent);
   color: var(--accent, var(--red));
@@ -20619,6 +20713,11 @@ body.gallery-selecting .gallery-dl-btn,
 }
 .cookbook-task[data-status="done"] .cookbook-task-check-ico { display: inline; }
 .cookbook-task[data-status="done"] .cookbook-task-clear-ico { display: none; }
+@media (max-width: 820px) {
+  .cookbook-task-check {
+    top: 2px;
+  }
+}
 .cookbook-task-start-now {
   display: inline-flex;
   align-items: center;
@@ -20652,24 +20751,30 @@ body.gallery-selecting .gallery-dl-btn,
 /* "Serve" button on a finished download — green pill matching the "running" /
    finished badge (it sits next to the green FINISHED chip + check). */
 .cookbook-task-serve-btn {
-  font-size: 9px;
+  display: inline-flex;
-  font-weight: 600;
+  align-items: center;
-  padding: 1px 6px;
+  gap: 3px;
-  border: none;
+  padding: 1px 6px 1px 4px;
-  border-radius: 3px;
+  border: 0;
-  line-height: 16px;
+  border-radius: 9px;
+  line-height: 1;
   flex-shrink: 0;
   cursor: pointer;
   font-family: inherit;
-  background: color-mix(in srgb, var(--green, #50fa7b) 20%, transparent);
+  font-size: 9px;
+  text-transform: lowercase;
+  background: transparent;
   color: var(--green, #50fa7b);
   position: relative;
   top: -2px;
+  margin-right: 2px;
   appearance: none;
   -webkit-appearance: none;
   -moz-appearance: none;
+  transition: background 0.15s;
 }
-.cookbook-task-serve-btn:hover { background: color-mix(in srgb, var(--green, #50fa7b) 32%, transparent); }
+.cookbook-task-serve-btn svg { flex-shrink: 0; }
+.cookbook-task-serve-btn:hover { background: color-mix(in srgb, var(--green, #50fa7b) 16%, transparent); }
 .cookbook-task-sub {
   padding: 1px 10px 4px;
   line-height: 1;
@@ -21448,6 +21553,31 @@ body.gallery-selecting .gallery-dl-btn,
 .cookbook-dl-btn:hover {
   opacity: 0.9;
 }
+.cookbook-dl-gguf-row {
+  margin-top: -1px;
+  gap: 5px;
+  align-items: center;
+  justify-content: flex-end;
+  font-size: 11px;
+  position: relative;
+  top: -2px;
+}
+.cookbook-dl-gguf-label {
+  opacity: 0.65;
+  flex-shrink: 0;
+}
+#cookbook-dl-gguf-quant {
+  height: 28px;
+  min-width: 118px;
+  flex: 0 0 auto;
+}
+#cookbook-dl-gguf-note {
+  opacity: 0.55;
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  max-width: 240px;
+}
 
 /* HF link in search panel */
 .hwfit-panel-hf-link {
@@ -31920,24 +32050,34 @@ body.notes-drag-mode .note-card-pin svg {
 .note-corner-menu-dropdown .ncm-item:hover {
   background: color-mix(in srgb, var(--fg) 8%, transparent);
 }
-/* "Agent" tag on a note that has a linked agent chat session */
+.note-checkbox-agent {
-.note-agent-tag {
-  align-self: flex-start;
   display: inline-flex;
   align-items: center;
-  gap: 5px;
+  justify-content: center;
-  background: color-mix(in srgb, var(--accent, var(--red)) 14%, transparent);
+  width: 14px;
-  border: 1px solid color-mix(in srgb, var(--accent, var(--red)) 35%, transparent);
+  height: 14px;
+  padding: 0;
+  margin: 0 1px;
+  border: 0;
+  background: transparent;
   color: var(--accent, var(--red));
-  border-radius: 999px;
+  box-shadow: none;
-  padding: 3px 10px 3px 8px;
-  font-size: 11px;
-  font-weight: 600;
   cursor: pointer;
-  margin-top: 2px;
+  opacity: 0;
-  transition: background 0.12s;
+  transition: opacity 0.12s, color 0.12s;
+}
+.note-checkbox:hover .note-checkbox-agent { opacity: 0.55; }
+.note-checkbox-agent:hover {
+  background: transparent;
+  opacity: 1 !important;
+}
+.note-checkbox-agent.is-agent-stream-complete {
+  color: #50fa7b;
+  opacity: 0.9;
+}
+.note-checkbox-agent svg {
+  display: block;
 }
-.note-agent-tag:hover { background: color-mix(in srgb, var(--accent, var(--red)) 24%, transparent); }
 
 .note-card {
   /* Same tint that .doclib-card uses so a default (uncolored) note
@@ -36414,6 +36554,10 @@ body.research-panel-view #research-divider { display:none; }
 .research-setting {
   display:flex; flex-direction:column; flex:1; min-width:90px;
 }
+.research-settings-row .research-setting:nth-last-child(-n + 3) {
+  position: relative;
+  top: 3px;
+}
 .research-setting-label {
   font-size:9px; text-transform:uppercase; letter-spacing:0.5px;
   opacity:0.5; margin-bottom:2px;

tests/test_fenced_example_not_executed_for_native_models.py

@@ -221,6 +221,60 @@ def test_skip_fenced_still_recovers_xml_invoke_markup():
     assert "latest python release" in blocks[0].content
 
 
+def test_stepfun_native_tool_tokens_are_executed_even_when_fenced_fallback_is_skipped():
+    leaked = (
+        "<｜tool▁calls▁begin｜>"
+        "<｜tool▁call▁begin｜>web_search<｜tool▁sep｜>"
+        '{"query":"Sweden news today"}'
+        "<｜tool▁call▁end｜>"
+        "<｜tool▁calls▁end｜>"
+    )
+    blocks = parse_tool_blocks(leaked, skip_fenced=True)
+    assert len(blocks) == 1
+    assert blocks[0].tool_type == "web_search"
+    assert "Sweden news today" in blocks[0].content
+    assert strip_tool_blocks(leaked, skip_fenced=True) == ""
+
+
+def test_stepfun_native_tool_tokens_accept_plain_web_query():
+    leaked = (
+        "<｜tool▁call▁begin｜>web_search<｜tool▁sep｜>"
+        "Sweden news today"
+        "<｜tool▁call▁end｜>"
+    )
+    blocks = parse_tool_blocks(leaked, skip_fenced=True)
+    assert len(blocks) == 1
+    assert blocks[0].tool_type == "web_search"
+    assert "Sweden news today" in blocks[0].content
+
+
+def test_skip_fenced_still_recovers_direct_xml_tool_markup():
+    leaked = (
+        "I'll search now.\n"
+        "<tool_call><web_search>News in Sweden today 2026-06-22</web_search></tool_call>"
+    )
+    blocks = parse_tool_blocks(leaked, skip_fenced=True)
+    assert len(blocks) == 1
+    assert blocks[0].tool_type == "web_search"
+    assert "News in Sweden today 2026-06-22" in blocks[0].content
+    assert strip_tool_blocks(leaked, skip_fenced=True) == "I'll search now."
+
+
+def test_skip_fenced_recovers_direct_xml_tool_markup_with_unclosed_wrapper():
+    leaked = (
+        "I'll search now.\n"
+        "<tool_call>\n"
+        "<web_search>\n"
+        "Sweden news today 2026-06-22\n"
+        "</web_search>"
+    )
+    blocks = parse_tool_blocks(leaked, skip_fenced=True)
+    assert len(blocks) == 1
+    assert blocks[0].tool_type == "web_search"
+    assert "Sweden news today 2026-06-22" in blocks[0].content
+    assert strip_tool_blocks(leaked, skip_fenced=True) == "I'll search now."
+
+
 def test_skip_fenced_still_recovers_dsml_markup():
     dsml = (
         "Let me search for that.\n"

tests/test_upload_multifile.py

@@ -19,7 +19,12 @@ from pathlib import Path
 
 import pytest
 from fastapi import APIRouter
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+from sqlalchemy.pool import NullPool
 
+import core.database as cdb
+from core.database import GalleryImage
 from src.upload_handler import count_recent_uploads, UploadHandler
 import routes.upload_routes as up
 
@@ -82,6 +87,10 @@ def _files(n):
     return [types.SimpleNamespace(filename=f"f{i}.txt") for i in range(n)]
 
 
+def _image_upload(name="photo.png", content=b"not really png but enough for route metadata"):
+    return types.SimpleNamespace(filename=name, file=io.BytesIO(content))
+
+
 @pytest.fixture(autouse=True)
 def _reset_router(monkeypatch):
     # Module-level router accumulates routes across setup calls; reset it.
@@ -163,3 +172,64 @@ def test_six_file_batch_is_not_rate_limited(tmp_path):
         assert meta and meta.get("id")
         saved += 1
     assert saved == 6
+
+
+async def test_chat_image_upload_is_added_to_gallery(tmp_path, monkeypatch):
+    engine = create_engine(
+        f"sqlite:///{tmp_path / 'gallery.db'}",
+        connect_args={"check_same_thread": False},
+        poolclass=NullPool,
+    )
+    cdb.Base.metadata.create_all(engine)
+    TestingSession = sessionmaker(bind=engine, autoflush=False, autocommit=False)
+    gallery_dir = tmp_path / "generated_images"
+
+    monkeypatch.setattr(up, "SessionLocal", TestingSession)
+    monkeypatch.setattr(up, "GENERATED_IMAGES_DIR", str(gallery_dir))
+
+    h = UploadHandler(base_dir=str(tmp_path), upload_dir=str(tmp_path / "uploads"))
+    up.setup_upload_routes(h)
+    endpoint = _endpoint(up.router)
+
+    result = await endpoint(_request(user="alice"), [_image_upload()])
+    uploaded = result["files"][0]
+
+    assert uploaded["gallery_id"]
+    db = TestingSession()
+    try:
+        image = db.query(GalleryImage).filter(GalleryImage.id == uploaded["gallery_id"]).one()
+        assert image.owner == "alice"
+        assert image.model == "chat-upload"
+        assert image.prompt == "photo.png"
+        assert image.file_hash == uploaded["hash"]
+        assert (gallery_dir / image.filename).exists()
+    finally:
+        db.close()
+
+
+async def test_non_image_chat_upload_is_not_added_to_gallery(tmp_path, monkeypatch):
+    engine = create_engine(
+        f"sqlite:///{tmp_path / 'gallery.db'}",
+        connect_args={"check_same_thread": False},
+        poolclass=NullPool,
+    )
+    cdb.Base.metadata.create_all(engine)
+    TestingSession = sessionmaker(bind=engine, autoflush=False, autocommit=False)
+    monkeypatch.setattr(up, "SessionLocal", TestingSession)
+    monkeypatch.setattr(up, "GENERATED_IMAGES_DIR", str(tmp_path / "generated_images"))
+
+    h = UploadHandler(base_dir=str(tmp_path), upload_dir=str(tmp_path / "uploads"))
+    up.setup_upload_routes(h)
+    endpoint = _endpoint(up.router)
+
+    result = await endpoint(_request(user="alice"), [types.SimpleNamespace(
+        filename="notes.txt",
+        file=io.BytesIO(b"plain text upload"),
+    )])
+
+    assert "gallery_id" not in result["files"][0]
+    db = TestingSession()
+    try:
+        assert db.query(GalleryImage).count() == 0
+    finally:
+        db.close()