LangGraph, CrewAI weigh local RAG

This is really an architecture question: all three can run on Ollama, so the win goes to whichever framework makes control flow, state, and debugging the least painful. LangGraph's state graph, checkpointing, and human-in-the-loop hooks fit retrieval, grading, and answer loops especially well. CrewAI wins on ergonomics, with flows, tracing, and an Ollama path via LiteLLM that is straightforward, but the abstraction is more opinionated. Microsoft Agent Framework looks promising for .NET/Azure teams and also supports Ollama, but Microsoft labels it public preview. For mixed 8B/32B/70B local models, the framework that makes routing, retries, and evaluation explicit will matter more than the number of agents.