Autogen with Streamlit

[cbsingh05iitkgp]

Sharing simple example for people who needed a sample to start with
https://github.com/cbsingh05iitkgp/AutoGenWithStreamLit

Code

def main():
    st.title("Chat System")

    if "messages" not in st.session_state:
        st.session_state.messages = []
        #st.session_state.last_message = None
        st.session_state.messages.append({"role": "assistant", "content": "How can I help you?"})
        #user_input=None;

    # Display chat messages
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])
    
    user_input = st.chat_input("Enter your message:")
    if user_input:
        # Append user input to messages
        st.session_state.messages.append({"role": "user", "content": user_input})
        with st.chat_message("user"):
            st.markdown(user_input)

        async def handle_message():
            # Create a task to send user input to the "team"
            task = HandoffMessage(
                source="user",
                target="Assistant",
                content=user_input
            )
            task_result = await team.run(task=task)
            # Get the last TextMessage
            last_message = next(
                (msg.content for msg in reversed(task_result.messages) if msg.type == "TextMessage"),
                None
            )

            # Append system response to messages
            st.session_state.messages.append({"role": "assistant", "content": last_message})
            #st.session_state.last_message = last_message

            with st.chat_message("assistant"):
                st.write(last_message)

        asyncio.run(handle_message())

[ekzhu]

Thanks! @cbsingh05iitkgp, would love to see this as a contribution to the python samples. You can take a look at the AgentChat with ChainLit sample. https://github.com/microsoft/autogen/tree/main/python/samples/agentchat_chainlit

Would you consider also make it possible to stream messages by using run_stream?