This project provides a solution for using GraphRAG with non-GPT large language models, targeting developers and researchers who need to build knowledge-based systems with flexible LLM integrations. It enables the use of various local and cloud-based LLMs as alternatives to proprietary models, offering cost-effectiveness and greater control over data processing.

How It Works

GraphRAG processes unstructured text to build a knowledge graph, extracting entities, relationships, and claims. It then uses community detection to create hierarchical summaries and embeds these into vector spaces. This structured data, combined with original text, powers different search strategies: local search for entity-specific queries and global search for broader dataset understanding. Prompt tuning allows for domain adaptation of the LLM interactions.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Requires Python 3.x.
• LLM API keys and configurations for chosen models (e.g., Ollama, Qwen, Gemini).
• Optional: Neo4j for knowledge graph visualization.
• Setup involves cloning the repository, installing dependencies, configuring .env and settings.yaml files with LLM details, and initializing the graphrag index.

Highlighted Details

• Supports a wide range of LLMs including Ollama, Qwen, Gemini, ChatGLM, SparkDesk, and Moonshot AI.
• Offers local and global search capabilities, plus question generation based on extracted entities.
• Includes Neo4j and 3D visualization for knowledge graphs.
• Supports incremental index updates and DRIFT graph reasoning search (version 0.4.0+).

Maintenance & Community

The project is based on Microsoft's GraphRAG (github.com/microsoft/graphrag). The NanGePlus/GraphragTest repository provides specific integration examples and guidance. Links to Bilibili video tutorials are provided for setup and advanced usage.

Licensing & Compatibility

The underlying GraphRAG project is licensed under the MIT License. This permissive license generally allows for commercial use and integration into closed-source projects.

Limitations & Caveats

The project's README focuses on specific integration examples and may require significant configuration for different LLMs or datasets. Performance and cost can vary greatly depending on the chosen LLM and usage patterns, as indicated by the provided token consumption examples. The project is actively being developed, with recent updates introducing significant structural changes.