Youtu-GraphRAG
StaleDescription
Open-source vertically unified agent GraphRAG framework from Tencent published at ICLR 2026. Integrates multi-agent collaboration with graph-based retrieval to tackle complex reasoning tasks on knowledge graphs.
Key Features
- Vertically unified architecture — integrates multi-agent collaboration with graph retrieval to avoid glue code
- Complex reasoning — designed for enterprise Q&A scenarios that require multi-hop reasoning
- ICLR 2026 paper — peer-reviewed academic foundation
- Multi-agent collaboration — sub-agents divide work across retrieval, reasoning, and answer generation
- Graph-driven — structured retrieval over knowledge graphs avoids the similarity-vs-relevance trap
- Open-source and reproducible — provides full training, inference, and evaluation scripts
Use Cases
💡 Building multi-hop Q&A over enterprise knowledge graphs
💡 Building explainable RAG systems where the retrieval path is traceable on the graph
💡 Replacing traditional vector RAG to avoid similar-but-irrelevant retrievals
💡 Reproducing and extending the ICLR 2026 GraphRAG paper
💡 Powering domain knowledge Q&A in finance, healthcare, and legal verticals
Strengths & Limitations
✅ Strengths
- • High community interest (1.2k stars)
⚠️ Limitations
- • No clear open-source license
Categories
Quick Start
git clone https://github.com/TencentCloudADP/youtu-graphrag.git
cd youtu-graphrag
pip install -r requirements.txt
export OPENAI_API_KEY=sk-...
python scripts/build_graph.py --input ./data --output ./graph
python scripts/agentic_rag.py --query "How does X affect Y?"