[Help Needed] Chainlit and autogen -- human-in-loop failed in the UI & unable to use code executor both in docker and locally

[irmathebest]

Hi folks. I have two questions related to the autogen and integration of chainlit and the selector group agent.
There are two bugs, I struggled a lot with.

1. the user input event stuck in the ui, whenever it asks me to get some input, I cannot put anything in the text box. It succesfully ran at the first time tho.
2. docker worked in jupyter notebook, not in chainlit ( it default image is python 3:slim but it always says the container is not running, I also tried image name one, it does not work also.... )
   Does anyone know how to resolve it? Thanks a lot!
   Here is my code:

@cl.on_chat_start
async def start_chat() -> None:
    

    code_executor = DockerCommandLineCodeExecutor( work_dir = "coding")
    a1 = AssistantAgent(
        "a1",
        description = "xxx, .",
        model_client=client,
        system_message="""xxxxx"""
    )
   
    user_agent = UserProxyAgent(
        "UserAgent",
        description = 'A human user to answer any questions, resolve conflict, provide confirmations, and be in loop for important decisions'
    )
    
    
    code_executor_agent = CodeExecutorAgent(
        "CodeExecutorAgent",
        code_executor=code_executor,
        description="""xxxxxxx""",
    )
    text_termination = TextMentionTermination("TERMINATE")
    max_messages_termination = MaxMessageTermination(max_messages=75)
    termination = text_termination | max_messages_termination
    selector_prompt = "xxxxxxxxx"
    group_chat = SelectorGroupChat(
        [a1,
        user_agent
        ,code_executor_agent], 
        model_client=client, 
        termination_condition=termination,
        selector_prompt=selector_prompt,
        allow_repeated_speaker=False
    )
 


    cl.user_session.set("prompt_history", "")  # type: ignore
    cl.user_session.set("team", group_chat)  # type: ignore
    await cl.Message(content="🤖 Everything is ready! Send a message to start.").send()


@cl.on_message  # type: ignore
async def chat(message: cl.Message) -> None:
    """
    Version 1
    """
    team = cast(SelectorGroupChat, cl.user_session.get("team"))  # type: ignore
    if team is None:
        print("❌ Error: Team is None!")
        await cl.Message(content="⚠️ Error: The system is not initialized correctly.").send()
        return

    print(f"📩 User Message: {message.content}")


    async for msg in team.run_stream(
        task=[TextMessage(content=message.content, source="UserAgent")],
        cancellation_token=CancellationToken(),
    ):
        if isinstance(msg, TextMessage):  
            print(f"📤 Agent {msg.source} replied: {msg.content}")
            await cl.Message(content=msg.content, author=msg.source).send()
        
        elif isinstance(msg, TaskResult):
            final_message = f"Task terminated. {msg.stop_reason or ''}"
            await cl.Message(content=final_message).send()
        else:
            await cl.Message(content="You are passing.").send()
            pass 

[ekzhu]

Have you seen this sample: https://github.com/microsoft/autogen/tree/main/python/samples/agentchat_chainlit

For code executor have you checked if the container is running use your docker desktop?

[irmathebest]

Yes I followed that example and I think there is some conflict between the code executor and the streaming thing so I switch to the current code. As you can see the structure is quite similar.
As for the code executor, when I run in Jupyter notebook it automatically pull the python image and create a new one. Also I checked that the container is running when I specify which container I am using. Thanks!

[ekzhu]

In your code I don't see how the code executor is created. Also, the user proxy doesn't have a custom input function. The default input function is console only.

[irmathebest]

Thank you so much, Eric let me try to add the user input function in my user proxy agent.
As for the code executor, I first add code_executor = DockerCommandLineCodeExecutor( work_dir = "coding")
and then I add the code executor agent
code_executor_agent = CodeExecutorAgent(
"CodeExecutorAgent",
code_executor=code_executor,
description="""xxxxxxx""",
)
not sure if there is anything else I need to change.

[ekzhu]

You need to call start on the executor. In notebook you probably used a context manager so it was automatically called.

https://microsoft.github.io/autogen/stable/reference/python/autogen_ext.code_executors.docker.html#autogen_ext.code_executors.docker.DockerCommandLineCodeExecutor.start

If it works, appreciate if you can submit a PR to fix the documentation of code executors to mention the start method

[irmathebest]

Perfect, thanks a lot. I will test it out and if it works, I can contribute to the documentation. As for the start method, do I need to do await, after the executor creation?