fix(llm): correct context sizing, honor think requests, broaden delete

Three related fixes uncovered while benchmarking qwen3:14b against 8b:

- pick_num_ctx was only counting message content, missing the ~15K
  tokens of tool schemas. num_ctx=8192 was being selected while actual
  prompt_tokens hit 14K+, causing silent prompt truncation on every
  tool-using request. Now includes json.dumps(tools) in the estimate.
  KV cache priming in app.py and routes/settings.py also fetches tools
  so the primed num_ctx matches what real chat requests will use.

- _should_think's heuristic classifier was overriding explicit
  think=true requests from the frontend toggle and MCP, gating on
  message length and regex patterns. Now a pass-through — the caller
  is the source of truth. quick_capture hardcodes think=False since
  it's a fast classification path that was relying on the old gating.

- delete_note description only mentioned "note or task", so the model
  refused to call it for entries created by save_person / save_place /
  create_list. Description now explicitly lists all five note_types it
  handles.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/fabledassistant/app.py

@@ -189,13 +189,16 @@ def create_app() -> Quart:
             """
             try:
                 from fabledassistant.services.llm import build_context, keep_alive_for, pick_num_ctx
+                from fabledassistant.services.tools import get_tools_for_user
                 messages, _ = await build_context(
                     user_id=user_id,
                     history=[],
                     current_note_id=None,
                     user_message=" ",
                 )
-                num_ctx = pick_num_ctx(messages)
+                # Include tool schemas so num_ctx matches real chat requests.
+                tools = await get_tools_for_user(user_id)
+                num_ctx = pick_num_ctx(messages, tools=tools)
                 async with httpx.AsyncClient(timeout=120.0) as client:
                     await client.post(
                         f"{Config.OLLAMA_URL}/api/chat",

src/fabledassistant/routes/quick_capture.py

@@ -11,7 +11,6 @@ from quart import Blueprint, jsonify, request
 
 from fabledassistant.auth import get_current_user_id, login_required
 from fabledassistant.config import Config
-from fabledassistant.services.generation_task import _should_think
 from fabledassistant.services.llm import stream_chat_with_tools
 from fabledassistant.services.tools import execute_tool, get_tools_for_user
 
@@ -53,11 +52,10 @@ async def quick_capture_route():
         {"role": "user", "content": text},
     ]
 
-    think = _should_think(text, think_requested=True)
-
+    # Quick capture is a fast classification path — never think.
     tool_calls: list[dict] = []
     try:
-        async for chunk in stream_chat_with_tools(messages, model, tools=capture_tools, think=think, num_ctx=4096):
+        async for chunk in stream_chat_with_tools(messages, model, tools=capture_tools, think=False, num_ctx=4096):
             if chunk.type == "tool_calls" and chunk.tool_calls:
                 tool_calls = chunk.tool_calls
     except Exception:

src/fabledassistant/routes/settings.py

@@ -16,6 +16,7 @@ async def _prime_kv_cache_bg(user_id: int, model: str) -> None:
     """Fire-and-forget: prime Ollama's KV cache with the user's system prompt."""
     import httpx
     from fabledassistant.services.llm import build_context, pick_num_ctx
+    from fabledassistant.services.tools import get_tools_for_user
     try:
         messages, _ = await build_context(
             user_id=user_id,
@@ -23,7 +24,11 @@ async def _prime_kv_cache_bg(user_id: int, model: str) -> None:
             current_note_id=None,
             user_message=" ",
         )
-        num_ctx = pick_num_ctx(messages)
+        # Size the prime to match what real chat requests will use, including
+        # tool schemas — otherwise Ollama reloads the model on the first real
+        # request and throws away the cache we just built.
+        tools = await get_tools_for_user(user_id)
+        num_ctx = pick_num_ctx(messages, tools=tools)
         from fabledassistant.services.llm import keep_alive_for
         async with httpx.AsyncClient(timeout=120.0) as client:
             await client.post(

src/fabledassistant/services/generation_task.py

@@ -37,62 +37,24 @@ _TOOL_CALL_MARKER = re.compile(r"^\s*\[TOOL_CALLS\]\s*", re.IGNORECASE)
 DB_FLUSH_INTERVAL = 5.0  # seconds between partial DB flushes
 
 # ---------------------------------------------------------------------------
-# Conditional thinking classifier
+# Thinking decision
 # ---------------------------------------------------------------------------
-
-# Patterns that force think=True even on short messages
-_THINK_FORCE = re.compile(
-    r"\b("
-    r"analyz|compar|explain\s+why|help\s+me\s+(think|plan|understand|figure\s+out)|"
-    r"step[- ]by[- ]step|debug|troubleshoot|diagnos|"
-    r"pros\s+and\s+cons|trade[- ]?off|"
-    r"architect|design\s+(a|the|my|this)|"
-    r"write\s+a\s+(detailed|long|comprehensive|full)|"
-    r"brainstorm|outline\s+(a|the|my)|"
-    r"what\s+(are|is)\s+the\s+(best|difference|relationship|impact|implication)|"
-    r"how\s+(do|does|should|would|can)\s+.{0,40}\s+work|"
-    r"why\s+(is|are|does|do|did|would|should)\b"
-    r")",
-    re.IGNORECASE,
-)
-
-# Patterns that force think=False regardless of message length
-_THINK_SKIP = re.compile(
-    r"^(hi|hey|hello|thanks|thank\s+you|ok|okay|got\s+it|sounds\s+good|"
-    r"great|perfect|sure|yes|no|yep|nope|nice|cool|awesome|"
-    r"what('s| is) \d|what time|how many|remind me|add (a |an )?(task|note|reminder)|"
-    r"create (a |an )?(task|note)|delete|update|mark .{0,30} (done|complete))\b",
-    re.IGNORECASE,
-)
-
-_WORD_COUNT_THRESHOLD = 60  # messages over this word count always use think=True
-_SHORT_MESSAGE_THRESHOLD = 12  # messages under this always use think=False
+#
+# The `think` flag from the frontend / MCP is taken at face value: if the
+# caller asked for thinking, they get thinking. No heuristic gating.
+#
+# Models that don't support extended reasoning (e.g. llama3, mistral) simply
+# ignore the `think` parameter in the Ollama chat request, so this is safe to
+# pass unconditionally across the full model zoo.
 
 
 def _should_think(user_content: str, think_requested: bool) -> bool:
     """Return whether extended thinking should be used for this request.
 
-    If the caller didn't request thinking, we never enable it. If they did,
-    we check whether the message is complex enough to warrant the overhead.
+    Honors the caller's request directly — no message-complexity classifier.
+    The frontend toggle / MCP `think` parameter is the source of truth.
     """
-    if not think_requested:
-        return False
-
-    text = user_content.strip()
-    word_count = len(text.split())
-
-    if word_count <= _SHORT_MESSAGE_THRESHOLD:
-        return False
-    if _THINK_SKIP.match(text):
-        return False
-    if word_count >= _WORD_COUNT_THRESHOLD:
-        return True
-    if "```" in text:
-        return True
-    if _THINK_FORCE.search(text):
-        return True
-
-    return False
+    return bool(think_requested)
 
 
 # Human-readable labels for each tool, shown in the status indicator
@@ -273,9 +235,10 @@ async def run_generation(
         voice_speech_style=voice_speech_style,
     )
 
-    # Pick the smallest context tier that fits the current messages.
+    # Pick the smallest context tier that fits the current messages AND the
+    # tool schemas (which can be 6-10K tokens on their own with ~40 tools).
     # Using the minimum needed tier reduces KV cache size and speeds up prefill.
-    num_ctx = pick_num_ctx(messages)
+    num_ctx = pick_num_ctx(messages, tools=tools)
     logger.debug("Adaptive num_ctx=%d for conv %d", num_ctx, conv_id)
 
     # Emit context event

src/fabledassistant/services/llm.py

@@ -38,12 +38,20 @@ def keep_alive_for(model: str) -> str:
     return Config.OLLAMA_KEEP_ALIVE_MAIN
 
 
-def pick_num_ctx(messages: list[dict]) -> int:
-    """Return the smallest context tier that fits *messages* with 25% headroom.
+def pick_num_ctx(messages: list[dict], tools: list[dict] | None = None) -> int:
+    """Return the smallest context tier that fits *messages* + *tools* with 25% headroom.
+
+    The ``tools`` JSON schemas are a large, often-overlooked chunk of the prompt.
+    With ~40 tools in the registry the schemas alone can be 6-10K tokens — enough
+    that omitting them from the estimate causes silent prompt truncation.
 
     Stays at or below Config.OLLAMA_NUM_CTX (the configured ceiling).
     """
     total_chars = sum(len(m.get("content") or "") for m in messages)
+    if tools:
+        # Serialize the same way Ollama will see them. json.dumps gives us a
+        # faithful char count for the schema payload without any guesswork.
+        total_chars += len(json.dumps(tools))
     estimated_tokens = int(total_chars / 3.5)
     needed = int(estimated_tokens * 1.25) + 256  # 25% headroom + output buffer
     cap = Config.OLLAMA_NUM_CTX

src/fabledassistant/services/tools/notes.py

@@ -391,9 +391,9 @@ async def list_notes_tool(*, user_id, arguments, **_ctx):
 
 @tool(
     name="delete_note",
-    description="Delete a note or task permanently. Use ONLY when the user explicitly asks to delete or remove an item. Always confirm with the user first — this cannot be undone.",
+    description="Delete any item from the user's knowledge base permanently — notes, tasks, persons (created via save_person), places (created via save_place), and lists (created via create_list) are all stored as notes and use this single delete tool. Use ONLY when the user explicitly asks to delete or remove an item. Always confirm with the user first — this cannot be undone.",
     parameters={
-        "query": {"type": "string", "description": "Title or keyword to find the note or task to delete"},
+        "query": {"type": "string", "description": "Title or keyword to find the item to delete (works for notes, tasks, persons, places, and lists)"},
         "confirmed": {"type": "boolean", "description": "Must be true — only set after the user has explicitly confirmed they want this item deleted."},
     },
     required=["query"],