Multi-agent functionality

[josecodes]

Hi @simonw! I'm thinking of building a thin multiagent layer on top of LLM. The layer would basically be something like two new tables to track agents (id, name) and multiagent conversations (conversation id, prompt agent id, response agent id), and some logic around that. It would be used via Python API; not sure what a CLI would look like.

Curious to hear your take on whether this sounds like a feature that would make sense to have in LLM OR as a plug-in perhaps OR separately on its own OR just a bad idea to try with LLM.

One of the things I really like about LLM is a built-in db for keeping prompts/responses for future use; this would be extra useful for multi-agent systems.