graphrag-local-ollama

Why we included this project

GraphRAG works well when you need to answer questions that span a whole document collection, but the official implementation leans on OpenAI models. This fork keeps the same two-stage pipeline, building an entity graph from your documents and precomputing community summaries, while swapping the model calls for anything you can pull through Ollama, embeddings included. It also ships a browser UI for people who'd rather not work from the command line. A LazyGraphRAG mode skips the summarization pass during indexing, which the project claims is about 99% faster, and generates those summaries on the fly when a query comes in; you trade a slower first answer for much cheaper indexing on large corpora. If you want to run GraphRAG-style retrieval entirely on your own hardware, without per-token cloud costs, this is a practical starting point.