Does autogen provide a way to manage multi-threads at the same time?

[Alive-Fish]

I hope to use autogen to simply our app (provide service for multi-users) and easily use the advanced features, but I found that autogen will manage intermediate results by itself. Also, it provides a way to save/load these results, but it might be a workaround.

Is there other better way for this? Or autogen is not designed for this scenario?

[jackgerrits]

In the core API this is managed via using separate topic sources and agent keys when publishing messages. This means that there are several agent instances, each with their own state. In agentchat I would use separate teams per user.

[Alive-Fish]

Thanks! It seems we should maintain the states for robustness whether using core API or agentchat.

Another question, is there a way to make agentchat stateless and call it with history which maintained by ourselves? The history only contains the user input and response of every turn. Does it a meaningful solution for service app?

[ekzhu]

Take a look at the agentchat fastapi sample. Saving and loading state can be achieved for agents and teams

https://github.com/microsoft/autogen/tree/main/python/samples/agentchat_fastapi