youtu-graphrag
Youtu-GraphRAG is a vertically unified agentic framework designed for Graph Retrieval-Augmented Generation, targeting complex reasoning tasks. Presented as an ICLR 2026 paper, this system integrates graph schema and retrieval into a single cohesive paradigm to enhance accuracy while reducing token costs by over 30 percent compared to state-of-the-art baselines. The software features a schema-guided hierarchical knowledge tree construction that supports four levels of abstraction: entity attributes, relation triples, keyword indexing, and community structures. It includes a novel dually-perceived community detection algorithm to identify relevant information clusters efficiently. Youtu-GraphRAG is built for seamless domain adaptation, allowing users to customize entity types and relations with minimal intervention, making it suitable for diverse applications such as encyclopedic search, academic paper analysis, and private enterprise knowledge bases. It excels in multi-hop reasoning, summarization, and handlin