feat: save account model and thinking settings for Hermes

agent/agent.py

@@ -111,9 +111,17 @@ async def entrypoint(ctx: JobContext):
             await stop_notice()
             notice_task = spawn(play_notice(text))
 
+    async def model_settings():
+        # Fetch at each turn, not just on join/poll: saved account changes apply
+        # immediately to the next request even in already-connected rooms.
+        current = await resolve_room(ctx.room.name)
+        if current["user"]["id"] != user_id:
+            raise RuntimeError("Account mismatch")
+        return current["model_settings"]
+
     bridge = HermesLLM(base_url=account["hermes_url"], api_key=account["hermes_api_key"], user_id=user_id,
                        room_id=ctx.room.name, on_progress=progress, on_wait_notice=wait_notice,
-                       on_answer_start=stop_notice)
+                       on_answer_start=stop_notice, get_model_settings=model_settings)
     tts = azure.TTS(voice=DEFAULT_VOICE, sample_rate=24000,
                     speech_key=os.getenv("AZURE_SPEECH_KEY", ""), speech_region=os.getenv("AZURE_SPEECH_REGION", "eastus"))
     session = AgentSession(stt=azure.STT(speech_key=os.getenv("AZURE_SPEECH_KEY", ""),

agent/hermes_llm.py

@@ -103,6 +103,7 @@ class HermesLLM(lk_llm.LLM):
         on_progress: ProgressCallback,
         on_wait_notice: WaitNoticeCallback | None = None,
         on_answer_start: Callable[[], Awaitable[None]] | None = None,
+        get_model_settings: Callable[[], Awaitable[dict[str, str]]] | None = None,
         timeout_seconds: float = 600.0,
         http_client: httpx.AsyncClient | None = None,
     ) -> None:
@@ -128,6 +129,7 @@ class HermesLLM(lk_llm.LLM):
         self._on_wait_notice = on_wait_notice
         self._last_wait_notice: str | None = None
         self._on_answer_start = on_answer_start
+        self._get_model_settings = get_model_settings
         self._timeout_seconds = timeout_seconds
         self._owns_client = http_client is None
         self._client = http_client if http_client is not None else httpx.AsyncClient(follow_redirects=False)
@@ -219,10 +221,25 @@ class HermesLLMStream(lk_llm.LLMStream):
             await asyncio.gather(timer, return_exceptions=True)
 
     async def _consume(self) -> None:
+        settings = {"model": "auto", "thinking_level": "none"}
+        if self._hermes._get_model_settings is not None:
+            try:
+                settings = await self._hermes._get_model_settings()
+            except Exception:
+                # Never fall back to stale preferences or run after revocation.
+                raise HermesLLMError("settings_unavailable") from None
+        model, effort = settings.get("model"), settings.get("thinking_level")
+        if (not isinstance(model, str) or not 1 <= len(model) <= 200
+                or not all(33 <= ord(c) <= 126 for c in model)
+                or effort not in ("none", "minimal", "low", "medium", "high", "xhigh")):
+            raise HermesLLMError("invalid_settings")
+        reasoning = {"enabled": False} if effort == "none" else {"enabled": True, "effort": effort}
         payload = {
-            "model": "hermes-agent", "stream": True,
+            # Explicit provider is required: Hermes otherwise ignores bare model
+            # requests by default. Neither provider nor API origin is client input.
+            "model": model, "provider": "custom:hope-upstream", "stream": True,
             "messages": _messages(self.chat_ctx),
-            "model_options": {"reasoning": {"enabled": False}},
+            "model_options": {"reasoning": reasoning},
         }
         finished = False
         async with self._hermes._client.stream(