How to Save and Resume Multi-Agent Conversation Sessions in Semantic Kernel?

[t-umemoto]

Hello.
I have a question regarding the multi-agent orchestration features in the Python version of Semantic Kernel (e.g., Magentic, GroupChat, Sequential, etc.).

In AutoGen, it is possible to save and restore the entire multi-agent conversation state (including message history and internal state) in JSON format using methods like save_state() / load_state(), and resume a session later.

I'd like to know whether similar session persistence is currently possible or planned in Semantic Kernel. Specifically:

1. Is it possible to save a conversation session (between the user and multiple agents) and later restore it to continue the conversation?
2. Is there a supported way to persist and reuse ChatHistory or ChatHistoryAgentThread within orchestrations?
3. Are there any current best practices or official plans regarding session persistence and resumption?

[moonbox3]

Hi @t-umemoto, thanks for the question.

1. We don't have a way (yet) to support saving/loading state. That'll be coming in the future.
2. As part of memory work, also coming, soon, we'll provide some methods to persist threads.
3. I think this part ties in within the first two above, and will follow naturally once the other two fundamental pieces are in place.

[t-umemoto]

Thank you for your response.

I understand that saving/loading state is not yet supported.

Being able to resume a session is a very important feature when developing AI applications.
I'll continue to keep an eye on this topic and I'm looking forward to future updates.