How to Catch Other Exceptions in AgentGroupChat?

[nblog]

Hi everyone,

I'm working with Python and semantic-kernel 1.22.1, and I'm using AgentGroupChat to handle chat interactions. Here is a simplified version of my code:

chat = AgentGroupChat()
await chat.add_chat_message(user_input)

try:
    async for message in chat.invoke(agent=agent):
        pass
except AgentChatException as e:
    raise
except Exception as e:
    raise

Problem:
It seems that only AgentChatException is raised, even when I expect errors like APIConnectionError, APITimeoutError, or other OpenAI-related exceptions.

Specifically, I noticed that when my OpenAI-compatible API server returns some custom response that is not a valid ChatCompletion, the following behavior occurs:

• The error originates from:

  semantic_kernel\connectors\ai\open_ai\services\open_ai_chat_completion_base.py
  

  
• Instead of receiving a more specific error, I only get:

  AgentChatException('Failed to select agent')

  
• This makes it difficult to catch the actual issue because all errors seem to be wrapped inside AgentChatException.

Question:

How can I properly capture and handle these underlying exceptions instead of only AgentChatException?
Is there a way to extract more detailed error information when the API response is invalid?

Any guidance would be greatly appreciated!

Thank you!

[markwallace-microsoft]

@alliscode Can you respond for this?

[moonbox3]

Adding @TaoChenOSU. Let's look how to improve this in the new agent chat abstractions.

[moonbox3]

Actually, @nblog, I've created an issue from this so we can track better to make improvements. Please follow along with #11369.