Human in loop integration with autogen multiagents

Using Streamlit
, , ,
1

I am trying to integrate the AutoGen Agents chatbot using the approach described in this article: Creating a Streamlit User Interface for AutoGen Agents | by Dr. Ernesto Lee | Medium.

the autogent userproxy code is look like this

class TrackableUserProxyAgent(UserProxyAgent):

def get_human_input(self, message):
    # Your custom function to handle human input
    print("Human input required:", message)
    
    user_query=""
    # Initialize the modal
    #modal = Modal(key="Provide input", title=message["content"], padding=20, max_width=744)
    # Define the content of the modal
    st.session_state['modal_open'] = True
    while st.session_state['modal_open']:
        st.stop()     

    # Additional content outside the modal
    user_query = st.session_state['dialog_input']
    print("User input:", user_query)
    st.write("This is the main app content.")
    
    # Return the human input (for example, from a user interface or other source)

    return user_query

def _process_received_message(self, message, sender, silent):
    #display_debug_messages(sender.name, message)
    return super()._process_received_message(message, sender, silent)

 # Initialize session state if needed
    if 'dialog_input' not in st.session_state:
        st.session_state['dialog_input'] = ""
    if 'modal_open' not in st.session_state:
        st.session_state['modal_open'] = False
        
    # Show modal until input is received
    #if st.session_state['modal_open']:
    self.show_dialog_popup(message)
        
    # Process input after modal is closed
    user_query = st.session_state['dialog_input']
    print("Received input:", user_query)
    return user_query

I have successfully implemented the interaction using the function def get_human_input(self, message):. However, whenever new answer is entered by user, the entire app refreshes, making it impossible to pass the input to the calling agents. I would appreciate any guidance on how to return the reply to the agents without refreshing the app