AgentGroupChat with AzureAssistantAgent agents and state management

[vslepakov]

Question: Reusing AzureAssistantAgent in AgentGroupChat

I have the following setup:

• An API that accepts user input
• It uses AgentGroupChat to orchestrate several AzureAssistantAgent instances
• Currently, I'm creating a new instance of AgentGroupChat on each request
• I attempt to reuse the AzureAssistantAgent instances by calling AzureAssistantAgent.retrieve with an agent_id (the agents are initially created on the first request)

Issue:
When I try to retrieve an existing agent by agent_id, I receive the error message: "No assistant found with id ..."

Questions:

• Is this error expected or by design?
• What is the correct approach for managing state and concurrency (multithreading) in AgentGroupChat when working with AzureAssistantAgent instances?

[moonbox3]

Hi @vslepakov, thanks for your question. Are you handling the retrieval in the following manner?

First we create the assistant agent, as an example:

agent = await AzureAssistantAgent.create(
    kernel=kernel,
    service_id=service_id,
    name=AGENT_NAME,
    instructions=AGENT_INSTRUCTIONS,
    enable_code_interpreter=True,
)

Then we can retrieve it using the assistant ID:

assistant_id = agent.assistant.id

# Retrieve the agent using the assistant_id
retrieved_agent: AzureAssistantAgent = await AzureAssistantAgent.retrieve(
    id=assistant_id,
    kernel=kernel,
)

# Define a thread and invoke the agent with the user input
thread_id = await retrieved_agent.create_thread()

try:
    await invoke_agent(retrieved_agent, thread_id, "Tell me a joke about bears.")
finally:
    await agent.delete()
    await retrieved_agent.delete_thread(thread_id)

Code is from this sample (I realize we have a bug in the sample where it only retrieves the OpenAIAssistantAgent, I will fix this):

https://github.com/microsoft/semantic-kernel/blob/main/python/samples/concepts/agents/assistant_agent_retrieval.py

[vslepakov]

Hi @moonbox3, thanks for your answer! Looking at your code, I just realized that I am retrieving the assistant_id like so agent.id instead of agent.assistant.id. Agent retrieval works now.

It would be great to have samples for agent usage with AgentGroupChat in the context of a web api (multiple simultaneous requests, agent and agent group chat lifecycle, history handling in AgentGroupChat when retrieving existing assistant agents etc.)

Thanks!