This repository provides a one-click deployment for a Knowledge Graph powered Retrieval Augmented Generation (RAG) system on Azure. It targets developers and researchers looking to leverage LLMs with enhanced context from knowledge graphs, offering a hosted API for indexing and querying. The solution aims to improve LLM outputs by integrating structured knowledge graph memory.

How It Works

The accelerator deploys a hosted API service on Azure, built upon the graphrag Python package. It enables automated indexing pipelines and querying of a knowledge graph. This approach enhances LLM responses by providing structured, contextual information derived from the knowledge graph, offering a more robust and informative interaction compared to standard RAG.

Quick Start & Requirements

• Deployment: Follow the deployment guide for a full API deployment.
• Prerequisites: Requires an Azure subscription. Deployment utilizes Bicep for provisioning multiple Azure services.
• Demonstration: A Quickstart notebook demonstrates API calls post-deployment.
• Cost Warning: Significant Azure costs may be incurred due to the use of multiple services and potential high utilization. Indexing operations can also be expensive.

Highlighted Details

• One-click deployment of a GraphRAG system on Azure.
• Exposes API endpoints for triggering indexing and querying.
• Leverages knowledge graph memory structures to enhance LLM outputs.
• Designed for a high-utilization API with auto-scaling and user access control.

Maintenance & Community

• This is a demonstration and not an officially supported Microsoft offering.
• Issues and feature requests are tracked via GitHub Issues.
• Contributions are welcome, subject to a Contributor License Agreement (CLA).
• Follows the Microsoft Open Source Code of Conduct.
• Roadmap and FAQ available at aka.ms/graphrag.

Licensing & Compatibility

• The repository's license is not explicitly stated in the provided README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The solution is a demonstration and not officially supported. It incurs substantial Azure costs and requires careful management of indexing operations due to expense. The README does not specify the exact license, which may impact commercial adoption.