Fix AI caller hanging on 'thinking...' indefinitely

- Add 30s timeout to all frontend fetch calls (safeFetch)
- Add 20s asyncio.timeout around lock+LLM in chat, ai-respond, auto-respond
- Reduce OpenRouter timeout from 60s to 25s
- Reduce Inworld TTS timeout from 60s to 25s
- Return graceful fallback responses on timeout instead of hanging

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

backend/main.py

@@ -1075,6 +1075,8 @@ async def chat(request: ChatRequest):
     session.add_message("user", request.text)
     session._research_task = asyncio.create_task(_background_research(request.text))
 
+    try:
+        async with asyncio.timeout(20):
             async with _ai_response_lock:
                 if _session_epoch != epoch:
                     raise HTTPException(409, "Call ended while waiting")
@@ -1094,6 +1096,9 @@ async def chat(request: ChatRequest):
                     messages=messages,
                     system_prompt=system_prompt
                 )
+    except TimeoutError:
+        caller_name = session.caller["name"] if session.caller else "Caller"
+        return {"text": "Uh... hold on, I lost my train of thought.", "caller": caller_name, "voice_id": session.caller["voice"] if session.caller else ""}
 
     # Discard if call changed while we were generating
     if _session_epoch != epoch:
@@ -1600,6 +1605,8 @@ async def _trigger_ai_auto_respond(accumulated_text: str):
 
     ai_name = session.caller["name"]
 
+    try:
+        async with asyncio.timeout(20):
             async with _ai_response_lock:
                 if _session_epoch != epoch:
                     return  # Call changed while waiting for lock
@@ -1620,6 +1627,10 @@ async def _trigger_ai_auto_respond(accumulated_text: str):
                     messages=messages,
                     system_prompt=system_prompt,
                 )
+    except TimeoutError:
+        print(f"[Auto-Respond] Timed out for {ai_name}")
+        broadcast_event("ai_done")
+        return
 
     # Discard if call changed during generation
     if _session_epoch != epoch:
@@ -1677,6 +1688,8 @@ async def ai_respond():
 
     epoch = _session_epoch
 
+    try:
+        async with asyncio.timeout(20):
             async with _ai_response_lock:
                 if _session_epoch != epoch:
                     raise HTTPException(409, "Call ended while waiting")
@@ -1694,6 +1707,8 @@ async def ai_respond():
                     messages=messages,
                     system_prompt=system_prompt
                 )
+    except TimeoutError:
+        return {"text": "Uh... sorry, I spaced out for a second there.", "caller": session.caller["name"], "voice_id": session.caller["voice"]}
 
     if _session_epoch != epoch:
         raise HTTPException(409, "Call changed during response")
@@ -1707,7 +1722,7 @@ async def ai_respond():
     ai_name = session.caller["name"]
     session.add_message(f"ai_caller:{ai_name}", response)
 
-    # TTS
+    # TTS — outside the lock so other requests aren't blocked
     audio_bytes = await generate_speech(response, session.caller["voice"], "none")
 
     if _session_epoch != epoch:

backend/services/llm.py

@@ -114,7 +114,7 @@ class LLMService:
         """Call OpenRouter API with retry"""
         for attempt in range(2):  # Try twice
             try:
-                async with httpx.AsyncClient(timeout=60.0) as client:
+                async with httpx.AsyncClient(timeout=25.0) as client:
                     response = await client.post(
                         "https://openrouter.ai/api/v1/chat/completions",
                         headers={

backend/services/tts.py

@@ -600,7 +600,7 @@ async def generate_speech_inworld(text: str, voice_id: str) -> tuple[np.ndarray,
         },
     }
 
-    async with httpx.AsyncClient(timeout=60.0) as client:
+    async with httpx.AsyncClient(timeout=25.0) as client:
         response = await client.post(url, json=payload, headers=headers)
         response.raise_for_status()
         data = response.json()

frontend/js/app.js

@@ -17,8 +17,12 @@ let sounds = [];
 
 
 // --- Safe JSON parsing ---
-async function safeFetch(url, options = {}) {
+async function safeFetch(url, options = {}, timeoutMs = 30000) {
-    const res = await fetch(url, options);
+    const controller = new AbortController();
+    const timer = setTimeout(() => controller.abort(), timeoutMs);
+    try {
+        const res = await fetch(url, { ...options, signal: controller.signal });
+        clearTimeout(timer);
         if (!res.ok) {
             const text = await res.text();
             let detail = text;
@@ -28,6 +32,11 @@ async function safeFetch(url, options = {}) {
         const text = await res.text();
         if (!text) return {};
         return JSON.parse(text);
+    } catch (err) {
+        clearTimeout(timer);
+        if (err.name === 'AbortError') throw new Error('Request timed out');
+        throw err;
+    }
 }