GraphRAG
MicrosoftMicrosoft Research's modular graph-based RAG system: LLMs extract a knowledge graph (entities + relationships) from unstructured text, cluster it into hierarchical communities, generate community summaries, and answer via Global/Local/DRIFT search - built for corpus-wide 'sensemaking' questions where naive vector RAG falls short.
MIT, no feature-gated core. ~34k stars. Indexing is compute/cost-intensive.
Openness
5 high confidence5.0
- license
- MIT(OSI)
- source
- public
- core-gated
- ungated
Fully MIT.
- https://github.com/microsoft/graphrag recorded 2026-06-17
MIT, ~33.8k stars
Adoption
3 medium confidence3.0
~34k stars; no published download telemetry - stars_fallback (capped at 3).
- https://github.com/microsoft/graphrag recorded 2026-06-17
~33.8k stars
Capability
4 medium confidence4.0
Turnkey for graph-RAG sensemaking but a deliberately narrow single technique.
- https://microsoft.github.io/graphrag/ recorded 2026-06-17
architecture + Global/Local/DRIFT modes
Unchanged since 2026-07-30 (last edited, not re-checked)