HKUDS/RAG-Anything: All-in-One Multimodal RAG Framework Surges to 16.4K Stars — 256 Stars Today

GitHub·medium signal

RAG-Anything built on LightRAG provides end-to-end multimodal document processing: PDFs, Office docs, images, tables, equations, and mixed content in a single pipeline. It automatically constructs knowledge graphs via entity extraction and cross-modal relationship discovery, then supports hybrid retrieval across textual and multimodal content. Recent additions include VLM-Enhanced Query mode for image-document integration. At 256 stars/day, it's addressing the gap where traditional RAG frameworks choke on non-text elements.