feat: add generation metrics to logs (think, rounds, tokens)

Log think flag, round count, prompt/output token counts per generation.
Change log category from 'usage' to 'generation' for clean MCP filtering.

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

src/fabledassistant/services/generation_task.py

@@ -212,7 +212,11 @@ async def run_generation(
 
     t_start = time.monotonic()
     timing: dict = {
+        "think": think,
         "tools": [],
+        "rounds": 0,
+        "prompt_tokens": None,
+        "output_tokens": None,
         "ttft_ms": None,
         "generation_ms": None,
         "total_ms": None,
@@ -228,6 +232,7 @@ async def run_generation(
         research_completed = False
 
         for _round in range(MAX_TOOL_ROUNDS + 1):
+            timing["rounds"] = _round + 1
             round_tool_calls: list[dict] = []
             logger.info("Generation round %d started for conv %d (model=%s)", _round, conv_id, model)
 
@@ -261,6 +266,12 @@ async def run_generation(
                             logger.warning("Failed periodic flush for message %d", msg_id, exc_info=True)
                         last_flush = now
 
+                elif chunk.type == "done":
+                    if chunk.prompt_tokens is not None:
+                        timing["prompt_tokens"] = (timing["prompt_tokens"] or 0) + chunk.prompt_tokens
+                    if chunk.output_tokens is not None:
+                        timing["output_tokens"] = (timing["output_tokens"] or 0) + chunk.output_tokens
+
                 elif chunk.type == "tool_calls" and chunk.tool_calls:
                     logger.info("Round %d: model returned %d tool call(s)", _round, len(chunk.tool_calls))
                     for tc in chunk.tool_calls:

src/fabledassistant/services/llm.py

@@ -149,6 +149,9 @@ class ChatChunk:
     type: Literal["content", "thinking", "tool_calls", "done"]
     content: str = ""
     tool_calls: list[dict] | None = None
+    # Token counts from the Ollama done event (only set on type="done")
+    prompt_tokens: int | None = None
+    output_tokens: int | None = None
 
 
 async def stream_chat_with_tools(
@@ -221,7 +224,11 @@ async def stream_chat_with_tools(
                         yield ChatChunk(type="tool_calls", tool_calls=accumulated_tool_calls)
                     else:
                         logger.debug("Ollama done with no tool calls")
-                    yield ChatChunk(type="done")
+                    yield ChatChunk(
+                        type="done",
+                        prompt_tokens=data.get("prompt_eval_count"),
+                        output_tokens=data.get("eval_count"),
+                    )
                     break
 
 

src/fabledassistant/services/logging.py

@@ -68,12 +68,12 @@ async def log_generation(
     """Persist per-generation timing breakdown to app_logs for benchmarking."""
     async with async_session() as session:
         log = AppLog(
-            category="usage",
+            category="generation",
             user_id=user_id,
             action="generation",
             endpoint=f"/chat/conversations/{conv_id}",
             duration_ms=timing.get("total_ms"),
-            details=json.dumps({"model": model, **timing}),
+            details=json.dumps({"model": model, "conv_id": conv_id, **timing}),
         )
         session.add(log)
         await session.commit()