# microsoft/graphrag

A modular graph-based Retrieval-Augmented Generation (RAG) system

Repository: https://github.com/microsoft/graphrag
Canonical: https://ross.abutalabs.com/products/graphrag
Homepage: https://microsoft.github.io/graphrag/
Language: Python
License: MIT
License Family: permissive
Topics: graphrag, rag, llm, llms, gpt, gpt-4, gpt4
Last push: 2026-08-24T18:30:25+00:00

## Health v2 (maintenance only)
Score: 91/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 63
- inputs: {"age_days": 889, "days_push": 9, "days_rel": 12, "gap_med": 14.5, "n_releases_24m": 35}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 35699, forks 3750 (observed 2026-08-28T04:12:00.903502+00:00)

## What it is
GraphRAG is a Python library and CLI pipeline from Microsoft Research that extracts a knowledge graph from unstructured text, builds community hierarchies with summaries, and uses those structures for graph-based retrieval-augmented generation. It improves LLM question answering over private datasets where naive vector-similarity RAG struggles to connect disparate information.

## Use cases
- build a knowledge graph from private documents for LLM question answering
- answer questions requiring synthesis across disparate sources in a corpus
- improve RAG quality over enterprise research or business documents
- generate hierarchical community summaries of a large text dataset
- run graph-based retrieval instead of vector similarity search

## When to choose
- your corpus is large and narrative, and baseline vector RAG fails to connect the dots
- you need global, dataset-level questions answered rather than snippet lookup
- you want a maintained, well-documented graph-based RAG pipeline with an indexer and query engine

## When to avoid
- you need active feature development or community PRs - the project is in maintenance mode
- you have a small budget - LLM-based indexing is expensive
- you only need simple semantic search over short documents - baseline RAG is cheaper and simpler
- you require an officially supported Microsoft product - this is a research demonstration

## Facets
- artifact type: library
- maturity: maintenance
- function: rag, llm-inference, etl, search-engine, nlp
- domain: large-language-models, artificial-intelligence
- platform: python, cli, cross-platform
- tags: knowledge-graph, graphrag, llm, gpt, question-answering, community-summarization, microsoft-research, retrieval-augmented-generation, natural-language-processing, data-engineering

## Member repositories
- microsoft/graphrag (main) score 91

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:12:00.903502+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:26:45.236366+00:00, confidence not recorded.
  - readme: https://github.com/microsoft/graphrag (fetched 2026-08-28T04:12:00.903502+00:00, sha d9363111a7ce)
  - homepage: https://microsoft.github.io/graphrag/ (fetched 2026-08-29T07:47:52.773333+00:00, sha 85e0c1fd8db8)
  - registry_pypi: https://pypi.org/pypi/graphrag/json (fetched 2026-08-29T07:47:52.782444+00:00, sha b60f8c744b80)
- Data as of 2026-08-30T08:39:29.467469+00:00.