Update app.py

app.py

@@ -13,7 +13,7 @@ from functools import lru_cache
 logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
 
 # -------------------- External Model Call (with Caching and Retry) --------------------
-@lru_cache(maxsize=128)  # Cache up to 128 responses
+# Removed @lru_cache here, as it caused issues with async and Gradio
 async def call_model(prompt: str, model: str = "gpt-4o", api_key: str = None, max_retries: int = 3) -> str:
     """Sends a prompt to the OpenAI API endpoint, with caching and retries."""
     if api_key is None:
@@ -344,20 +344,20 @@ def get_human_feedback(placeholder_text):
     """Gets human input using a Gradio Textbox."""
     with gr.Blocks() as human_feedback_interface:
         with gr.Row():
-            human_input = gr.Textbox(lines=4, label="Human Feedback", placeholder=placeholder_text) #Removed placeholder
+            human_input = gr.Textbox(lines=4, label="Human Feedback", placeholder=placeholder_text)
         with gr.Row():
             submit_button = gr.Button("Submit Feedback")
 
-        feedback_queue = queue.Queue()
-
         def submit_feedback(input_text):
-            feedback_queue.put(input_text)
-            return ""
+            # Put the feedback into the shared queue
+            human_input_queue.put(input_text)
+            return ""  # Clear the input box after submission
+
 
-        submit_button.click(submit_feedback, inputs=human_input, outputs=human_input)
+        submit_button.click(fn=submit_feedback, inputs=human_input, outputs=human_input)
         human_feedback_interface.load(None, [], [])  # Keep interface alive
 
-    return human_feedback_interface, feedback_queue
+    return human_feedback_interface
 
 # -------------------- Chat Function for Gradio --------------------
 
@@ -368,28 +368,53 @@ def multi_agent_chat(message: str, history: List[Any], openai_api_key: str = Non
         if not openai_api_key:
             yield "Error: API key not provided."
             return
+
     human_in_the_loop_event = threading.Event()
-    human_input_queue = queue.Queue() #For receiving the feedback request
+    human_input_queue = queue.Queue()  # Use a single queue for both requests and responses
+
+    # Start the conversation in a separate thread
+    conversation_thread = threading.Thread(
+        target=lambda: asyncio.run(
+            multi_agent_conversation(message, queue.Queue(), openai_api_key, human_in_the_loop_event, human_input_queue)
+        )
+    )
+    conversation_thread.start()
+
 
-    yield from process_conversation_generator(message, openai_api_key, human_in_the_loop_event, human_input_queue)
+    log_queue = queue.Queue()  # Local log queue for this chat instance
+    asyncio.run(multi_agent_conversation(message, log_queue, openai_api_key, human_in_the_loop_event, human_input_queue))
 
-    while human_in_the_loop_event.is_set():
-        yield "Waiting for human feedback..."
+
+
+    while conversation_thread.is_alive() or not log_queue.empty() or human_in_the_loop_event.is_set():
+        # Yield log messages
         try:
-          feedback_request = human_input_queue.get(timeout=0.1) #Non-blocking, check for feedback request
+            log_message = log_queue.get_nowait()  # Non-blocking get
+            if isinstance(log_message, tuple) and log_message[0] == "result":
+                final_result_text = "\n=== Conversation ===\n"
+                for entry in log_message[1]:
+                  final_result_text+= f"[{entry['agent']}]: {entry['message']}\n\n"
+                yield final_result_text
+            else:
+              yield log_message
+        except queue.Empty:
+            pass
+
+        # Handle human feedback requests
+        if human_in_the_loop_event.is_set():
+            yield "Waiting for human feedback..."
+            try:
+                feedback_request = human_input_queue.get(timeout=0.1) # Get context for the request.
+                human_interface = get_human_feedback(feedback_request) #Show context in the input box
+                yield gr.Textbox.update(visible=False), gr.update(visible=True) # Show the human feedback interface.
+                human_in_the_loop_event.wait() # Wait until the event is cleared
 
-          human_interface, feedback_queue = get_human_feedback(feedback_request)
+            except queue.Empty:
+                pass
+        await asyncio.sleep(0.1)  # Prevent busy-waiting
 
-          #This is a hacky but currently only working way to make this work with gradio
-          yield gr.Textbox.update(visible=False), gr.update(visible=True)
-          human_feedback = feedback_queue.get(timeout=300)  # Wait for up to 5 minutes
-          human_input_queue.put(human_feedback) #Put feedback where Orchestrator can find it.
-          human_in_the_loop_event.clear()
-          yield gr.Textbox.update(visible=True), human_interface.close() #Hide human input box
-          yield from process_conversation_generator(message, openai_api_key, human_in_the_loop_event, human_input_queue)
+    conversation_thread.join()
 
-        except queue.Empty: #If we get here, there was NO human feedback request, so skip.
-          continue #Go back to the top of the while loop
 
 # -------------------- Launch the Chatbot --------------------