The GraphRAG Toolkit is a Python library for creating Generative AI applications that leverage knowledge graphs. It targets developers and researchers building complex question-answering systems, offering a structured approach to integrate LLMs with graph data for enhanced accuracy and context.

How It Works

The toolkit offers two primary components: Lexical Graph and BYOKG-RAG. Lexical Graph automates the creation of hierarchical graphs from unstructured data, enabling question-answering strategies that query this graph. BYOKG-RAG facilitates Knowledge Graph Question Answering (KGQA) by combining LLMs with user-provided knowledge graphs for sophisticated querying.

Quick Start & Requirements

Installation instructions and requirements are detailed separately for each tool within the toolkit.

Highlighted Details

• Supports Neo4j graph stores for lexical graphs.
• Includes performance improvements for traversal-based retrievers.
• Offers a separate BYOKG-RAG package for integrating custom knowledge graphs.
• Features an MCP server for dynamic tool generation in multi-tenant environments.

Maintenance & Community

The project is part of AWS Labs. Further community and contribution details are available via the CONTRIBUTING guide.

Licensing & Compatibility

This project is licensed under the Apache-2.0 License, permitting commercial use and integration with closed-source applications.

Limitations & Caveats

FalkorDB support has been moved to a separate lexical-graph-contrib package, requiring separate installation for users who need it.