microsoft
ThoughtEvent fails to be transmitted to the AutoGen Studio UI
What happened?
Describe the bug Thought events are not transmitted to the UI in a swarm team with three agents. The intermediate responses are skipped before each handoff.
Possible causes:
- The ThoughtEvent class is not imported in the connection.py
- And the ThoughtEvent is filtered out in a subsequent isinstance check.
To Reproduce Create a three-agent swarm team with agents Alice, Bob, and Charlie without any function tool call.
The system message for Alice: "You are Alice, a thoughtful analyst who provides the initial perspective on any topic. When you receive a question or topic from the user, provide a comprehensive initial analysis with key points and insights. After sharing your perspective, always transfer the conversation to Bob using the transfer_to_bob function so he can add his viewpoint."
The system message for Bob: "You are Bob, an evaluator who adds a second perspective and directs the conversation flow. When you receive input from Alice, provide your own analysis and perspective on the topic. Then make a decision: If the topic needs more back-and-forth discussion, use transfer_to_alice to continue the dialogue. If the topic is ready for final synthesis and conclusion, use transfer_to_charlie. Make your decision based on the complexity and completeness of the discussion so far."
The system message for Charlie: "You are Charlie, the final synthesizer who brings the conversation to a conclusion. When you receive the discussion from Bob, provide a comprehensive synthesis that combines Alice's and Bob's perspectives, draws final conclusions, and summarizes key takeaways. End your response with 'TERMINATE' to signal the completion of the discussion."
Expected behavior The intermediate messages (for example, the conversation between Alice and Bob) are not displayed in the UI.
Which packages was the bug in?
AutoGen Studio (autogensudio)
AutoGen library version.
Python dev (main branch)
Other library version.
No response
Model used
No response
Model provider
None
Other model provider
No response
Python version
None
.NET version
None
Operating system
None
Hi @libinruan ,
Thanks for this. What model are you uising to test this? Can you share a sample config in python or serialized json for reproducibility?
