Managing Conversation History with LangGraph Supervisor?

[chrissimmerman]

I have created a multi agent architecture using the prebuilt create_supervisor() function in langgraph-supervisor. I noticed that there's no prebuilt way to manage conversation history within the supervisor graph, which means there's nothing that can be done when the context window length exceeds because of too many messages in the conversation.

Has anyone implemented a way to manage conversation history with langgraph-supervisor?

[chrissimmerman]

Looks like the best way to do this is use a pre_model_hook
https://langchain-ai.github.io/langgraph/agents/memory/#summarize-message-history