GraphRAG
MicrosoftMicrosoft Research's graph-based RAG system. A model extracts entities and relationships from unstructured text into a knowledge graph, clusters it into hierarchical communities, generates community summaries, and answers through global, local and DRIFT search modes. It targets corpus-wide sensemaking questions that vector retrieval alone handles poorly.
Indexing is compute- and cost-intensive relative to a vector pipeline. Verified 2026-08-13 via the microsoft/graphrag repository and its documentation site.
Openness
5 high confidence- license
- MIT(OSI)
- source
- public
- core-gated
- ungated
Fully MIT: the repo publishes the complete indexing and query pipeline and offers no hosted tier, enterprise directory or license key, so the core is ungated as well as OSI-licensed.
- https://github.com/microsoft/graphrag recorded 2026-08-13
MIT license, full public source for the indexing and query pipeline, 35.5k stars and 3.7k forks; no hosted tier or enterprise directory
Adoption
2 high confidenceThe pypistats API answers 69,377 downloads in the last month for the declared PyPI package graphrag (13,302 in the last week, 2,016 in the last day), which bands at 10K-100K, level 2 on the software scale. A download count outranks a star count, so the roughly 35.5k stars recorded on the openness axis corroborate the band rather than carry it.
- https://pypistats.org/api/packages/graphrag/recent recorded 2026-08-14
last_month = 69,377 downloads for graphrag (last_week 13,302, last_day 2,016).
Capability
4 medium confidenceTurnkey for graph-RAG sensemaking but a deliberately narrow single technique. No benchmark placement exists for a RAG technique, so this stays a feature-matrix judgment, one below the openhands anchor.
- https://microsoft.github.io/graphrag/ recorded 2026-08-13
architecture, knowledge-graph extraction, community detection and summarization, Global/Local/DRIFT query modes; no benchmark figure
Verified 2026-08-13