Add some kind of mechanism/component for managing state between Agent calls

[Finndersen]

Love the framework! I've used it to build an AI agent app, and needed to create an AgentRunner wrapper class to manage state (e.g. message history) between agent calls, and there's other useful functionality that could be added at this layer:

• inspecting/clearing history
• automatic pruning / compression of message history based on context limits as per Message history is too long with OpenAI #619).

Would something like this be a worthwhile addition to the framework?

[samuelcolvin]

we need to implement #619, we also need support for persistence see #695, I don't think we need anything more.

[Finndersen]

#695 appears to be for long term DB persistence between app runs/interactions, I'm referring to a helper for maintaining message history between multiple agent.run() calls within a single app execution/interaction (that might not need to be persisted afterwards), to facilitate back-and-forth flow with user.

Maybe this depends a bit on the context in which the application is running. For a web-server based API, persistence is required between each run because each request is independent. But for an interactive CLI style interface, there may be multiple runs within a single app invocation, without need for longer term persistence (like my AgentRunner class).

I guess an in-memory implementation of #695 might work