Update app.py

app.py

@@ -37,15 +37,31 @@ async def chat(request: Request, background_tasks: BackgroundTasks):
         user_input = data["user_input"]
         mode = data["mode"]
 
-        async def stream_response():
-            async for chunk in process_request(user_input, mode):
+        async def response_generator():
+            future = asyncio.Future()
+            request_queue.put((future, user_input, mode))
+            await future
+
+            async for chunk in future.result():
                 yield chunk
 
-        return StreamingResponse(stream_response(), media_type="text/plain")
+        return StreamingResponse(response_generator(), media_type="text/plain")
     except Exception as e:
         print(f"[SYSTEM] Error: {str(e)}")
         return {"error": str(e)}
 
+async def queue_worker():
+    while True:
+        if not request_queue.empty():
+            future, user_input, mode = request_queue.get()
+            result = process_request(user_input, mode)
+            future.set_result(result)
+        await asyncio.sleep(0.1)
+
+@app.on_event("startup")
+async def startup_event():
+    asyncio.create_task(queue_worker())
+
 @app.post("/sleep")
 async def sleep():
     try: