Repolode
Submit
microsoft

microsoft/graphrag

A Python data pipeline that uses LLMs to extract structured knowledge graphs from unstructured text for retrieval-augmented question answering.

View on GitHubWebsite
Stars
36,391
Forks
3,861
Open issues
57
Last commit
yesterday

What is it?

GraphRAG is a modular, graph-based retrieval-augmented generation system from Microsoft Research. It is a data pipeline and transformation suite that uses LLMs to pull structured data, such as entities and relationships, out of unstructured text, and then organizes that data into a graph that can supply targeted context for answering questions over private documents.

The project targets questions about large collections of narrative text, where a plain vector search may miss themes that span many documents. Indexing is the costly step: the project's guidance recommends reading the documentation on the process and costs involved and starting with a small dataset. Prompts generally need tuning for a specific dataset before results are good.

The project is a research effort now in maintenance mode, so it accepts no new features or PRs and only receives bug fixes, dependency updates and security patches. Its code is a demonstration rather than an officially supported Microsoft product.

Key features

  • 01Uses LLMs to extract entities and relationships from unstructured text into structured data.
  • 02Builds a knowledge graph from extracted data to provide context for question answering.
  • 03Provides a command line interface with an init step for generating configuration and prompts.
  • 04Includes a prompt tuning workflow for adapting extraction prompts to your data.
  • 05Distributed as a Python package on PyPI.

Use cases

  • Answering thematic questions over a large private corpus of narrative documents.
  • Exploring relationships among people, organizations or events in unstructured text.
  • Experimenting with graph-based RAG approaches in research settings.

Quick start

  1. 1

    Follow the command line quickstart in the official docs

  2. 2

    Initialize a project folder with the config format

    graphrag init --root [path] --force

Things to know

  • The project is in maintenance mode and is not accepting new PRs or features.
  • GraphRAG indexing can be expensive, so start with a small dataset.
  • The code is a research demonstration and not an officially supported Microsoft product.
  • Out-of-the-box results may be weak, so prompts usually need tuning.

Topics

#rag#llm#knowledge-graph#graph#retrieval#python#question-answering#gpt#gpt-4#gpt4

FAQ

What is graphrag?
A Python data pipeline that uses LLMs to extract structured knowledge graphs from unstructured text for retrieval-augmented question answering. GraphRAG is a modular, graph-based retrieval-augmented generation system from Microsoft Research. It is a data pipeline and transformation suite that uses LLMs to pull structured data, such as entities and relationships, out of unstructured text, and then organizes that data into a graph that can supply targeted context for answering questions over private documents.
Is graphrag free to use? What license does it have?
graphrag is open source under the MIT license. Check the license terms before using it in commercial products.
Is graphrag still maintained?
Actively maintained. The last commit to graphrag was on Oct 19, 2026.
What can I use graphrag for?
Answering thematic questions over a large private corpus of narrative documents. Exploring relationships among people, organizations or events in unstructured text. Experimenting with graph-based RAG approaches in research settings.
How do I install graphrag?
Follow the command line quickstart in the official docs Initialize a project folder with the config format: graphrag init --root [path] --force
Where can I get graphrag?
The source code and installation instructions are on GitHub: https://github.com/microsoft/graphrag

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects