Autogen_GraphRAG_Ollama

Why we included this project

Teams that want a fully offline question-answering stack over their own documents will find this a practical reference build. It wires Microsoft's GraphRAG knowledge-graph search into AutoGen agents, so a query can trigger function calls that pull from the graph rather than relying on a single vector lookup. Everything runs through Ollama for local inference and embeddings, with a Lite-LLM proxy bridging AutoGen's function-calling expectations to non-OpenAI models, and a Chainlit front end for continuous chat sessions. The setup is more involved than a one-command install: you patch a couple of GraphRAG package files and run several services, but the README walks through both Linux and Windows paths step by step. It is best treated as a working blueprint for assembling local agentic RAG, especially if you want to avoid any cloud API dependency while still getting graph-based retrieval.