GraphRAG

Overview

GraphRAG is a structured, hierarchical approach to Retrieval Augmented Generation. It extracts a knowledge graph from raw text, builds a community hierarchy, generates summaries for these communities, and leverages these structures for RAG-based tasks.

Key Features

• Extracts entities, relationships, and key claims from TextUnits.
• Builds a hierarchical community structure using the Leiden technique.
• Generates bottom-up summaries of each community and its constituents.
• Supports Global, Local, DRIFT, and Basic query modes.
• Provides prompt tuning guidance for fine-tuning prompts to your data.
• Offers CLI and Indexer/Query packages for working with the system.

Use Cases

• Reasoning over private datasets such as enterprise proprietary research, business documents, or communications.
• Answering holistic questions over large data collections or singular large documents.
• Connecting disparate pieces of information through shared attributes to synthesize new insights.
• Question answering about specific entities by fanning out to their neighbors and associated concepts.

Getting Started

• Check out the Get Started guide.
• Try the command line quickstart.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• GraphRAG is described as a research project that is largely in maintenance mode and will not accept new PRs or implement new features.
• The provided code is a demonstration and is not an officially supported Microsoft offering.
• Indexing can be expensive, so users should read the documentation to understand costs and start small.