This project integrates Microsoft's GraphRAG technology into Open WebUI, offering a versatile information retrieval API for advanced Q&A systems and search engines. It targets developers and users seeking to enhance web interfaces with sophisticated, graph-based search capabilities, combining local, global, and web searches for more comprehensive results.

How It Works

The system leverages GraphRAG's graph-based retrieval methods to improve search accuracy and relevance. It supports three primary retrieval strategies: local search for structured knowledge bases, global search for broader context, and Tavily search for external web information. A "Full Model Search" option combines all three, automatically integrating and ranking results from diverse sources.

Quick Start & Requirements

• Install: Clone the repository, create a virtual environment, and run pip install -r requirements.txt.
• Prerequisites: Python 3.8+, Tavily API key, and potentially local LLM/embedding models (Ollama, LM Studio).
• Configuration: Set environment variables for API keys (Tavily, LLM, embedding), model names, and API base URLs.
• Usage: Start the server with python main-en.py. Open WebUI can be configured to use the API endpoint http://localhost:8012/v1/chat/completions.
• Docs: Video Demo

Highlighted Details

• Supports local LLMs and embedding models via Ollama and LM Studio for enhanced privacy and cost reduction.
• Integrates local, global, and Tavily web search capabilities.
• Offers a "full-model" search that combines all retrieval methods.
• Provides an API endpoint compatible with Open WebUI.

Maintenance & Community

The project is maintained by win4r. Contact information for issues is provided via WeChat (stoeng). Contributions via pull requests are welcome.

Licensing & Compatibility

• License: Apache-2.0 License.
• Compatibility: Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

The graphrag package installation might require specific instructions from Microsoft Research if the standard pip install fails. The project relies on asynchronous programming, requiring compatible environments. For production, an ASGI server is recommended.