GraphRAG
A Microsoft research project and data pipeline that extracts structured, graph-based context from unstructured text using LLMs, to help a language model reason over private data.
GraphRAG builds a knowledge-graph memory structure out of a body of text and uses it to form targeted context for question answering, as an alternative to plain vector-based retrieval. It ships as a command-line quickstart along with configuration for prompt tuning, since results on a new dataset are unlikely to be good without adjusting prompts to that data; a migration path and notebook are provided for moving between versions without losing prior indexing work.
The project describes itself as a demonstration of the underlying methodology rather than an officially supported Microsoft product, and states it is now largely in maintenance mode: it takes bug fixes and dependency or security updates but is not accepting new features or pull requests, since Microsoft's research in this area has since diversified into other projects.
It is written in Python and licensed under the MIT license.