The Graphrag Toolkit from Awslabs is a Python-based solution designed for building graph-enhanced Generative AI applications, particularly Retrieval Augmented Generation (RAG) workflows. It supports Python versions 3.10-3.12 and is distributed via PyPI using "pip install graphrag" or "pip install graphrag-toolkit". This toolkit helps developers and ML engineering teams transform unstructured data into intelligent, relationship-aware knowledge systems. Its core purpose is to automate knowledge graph construction and compose effective question-answering strategies, integrating with various external services for data storage and AI capabilities.

What does Graphrag Toolkit connect to?

The Graphrag Toolkit integrates with various external systems and services. It connects to graph databases such as Amazon Neptune (Neptune Database, Neptune Analytics), FalkorDB, and Neo4j. For vector storage, it uses Amazon OpenSearch Serverless. Its AI capabilities come from Large Language Models and Foundation Models via Amazon Bedrock, and it can also configure OpenAI or Azure OpenAI models. The toolkit's category is specifically for building graph-enhanced Generative AI applications, focusing on Retrieval Augmented Generation workflows.

Core features and tools

The toolkit provides several callable tools and components. Command-line interface tools include "graphrag init" for workspace setup and "graphrag index" for running the indexing pipeline. A central component is the "lexical-graph," offering a framework for automating graph construction and developing question-answering strategies. It also features an EntityLinker component, which aligns natural language terms to structured data schemas. For graph store integration, methods like GraphStoreFactory.register() and GraphStoreFactory.for_graph_store() are exposed.

Who is it for?

Graphrag Toolkit serves developers building Generative AI applications, ML engineering teams, and data scientists. Organizations aiming to transform unstructured data into intelligent, relationship-aware knowledge systems will find it valuable. It supports complex question answering over knowledge graphs, enabling LLMs to reason across multiple documents. This is not the right fit if your project does not involve graph databases or advanced knowledge graph construction from unstructured text.

How to get started

The server authenticates using API keys for external LLM services, typically stored in an .env file, or via AWS credentials and execution roles for AWS services. It can also use Azure Managed Identity.

Use cases

• Building graph-enhanced Retrieval Augmented Generation (RAG) applications by automating knowledge graph construction with vector embeddings.
• Performing complex question answering over knowledge graphs, allowing LLMs to reason across multiple documents and relationships.
• Transforming unstructured and semi-structured textual content into structured knowledge graphs through entity extraction and relationship identification.

The Graphrag Toolkit offers a specialized approach to building Generative AI applications by integrating knowledge graphs. It provides essential components for automating graph creation from diverse data sources and enhancing RAG workflows. The toolkit's compatibility with various databases and AI models makes it a versatile option for complex data challenges. Try initializing a new workspace with "graphrag init" to explore its configuration options.