GraphRAG

GraphRAG

GraphRAG is Microsoft Research's graph-based retrieval-augmented generation system. Where conventional RAG retrieves isolated text chunks, GraphRAG builds a knowledge graph of entities and relationships from your documents, enabling LLMs to reason across the entire corpus and answer complex, multi-hop questions that vector search cannot.

Key Features

• Graph indexing — extracts entities, relationships, and community summaries from documents using an LLM
• Global search — answer questions that require synthesizing information across the entire document set
• Local search — entity-anchored retrieval for specific, focused questions
• Community reports — hierarchical cluster summaries provide high-level corpus understanding
• DRIFT search — Dynamic Reasoning and Inference with Flexible Traversal for hybrid global/local queries
• Prompt tuning — auto-generate domain-adapted extraction prompts from a sample of your data

Quick Start

pip install graphrag

# Initialize and index a corpus
mkdir -p ./rag/input && cp my_docs/*.txt ./rag/input/
graphrag init --root ./rag
# Configure ./rag/settings.yaml with your OpenAI API key
graphrag index --root ./rag

# Query
graphrag query --root ./rag --method global --query "What are the main themes?"

Install via ai-supply

npx ai-supply add graphrag-knowledge-graph-rag

Curated mirror of the open-source GraphRAG project (MIT). Install upstream from the repository.