Graph RAG Agent

Stale
GitHub Python MIT

Description

An intelligent RAG system integrating GraphRAG, LightRAG, and Neo4j graph builders with DeepSearch for reasoning-augmented retrieval and a custom evaluation framework for GraphRAG.

Key Features

  • Complete GraphRAG implementation with knowledge graph representation for reasoning-augmented retrieval
  • DeepSearch and GraphRAG fusion combining multi-step think-search-reason paradigm with graph structure
  • Multi-agent architecture with NaiveRagAgent, GraphAgent, HybridAgent, DeepResearchAgent, and FusionGraphRAGAgent
  • Plan-Execute-Report multi-agent collaboration with Planner, WorkerCoordinator, and Reporter modules
  • Incremental graph updates with entity disambiguation and alignment for knowledge graph maintenance
  • Comprehensive evaluation framework with 20+ metrics for answer quality, retrieval performance, and graph assessment

Use Cases

💡 Building enterprise-grade knowledge graph Q&A systems with explainable reasoning
💡 Deep research tasks requiring multi-step reasoning over private document collections
💡 Constructing and maintaining knowledge graphs from multi-format documents (PDF, DOCX, CSV, JSON)
💡 Evaluating and benchmarking GraphRAG-based retrieval systems with standardized metrics
💡 Visualizing AI reasoning traces for debugging and transparency in complex queries

Tags

graph-rag knowledge-graph neo4j deep-search evaluation

Categories

📚 RAG Tools

Quick Start

1. Clone repo: git clone https://github.com/1517005260/graph-rag-agent.git
2. Install dependencies and configure .env with Neo4j and LLM API keys
3. Build knowledge graph from documents using the pipeline modules
4. Run the FastAPI server (server/) or Gradio frontend (frontend/) for interactive Q&A
5. Test with: cd test && python search_with_stream.py
Visit GitHub

Related Projects