LightRAG

Graph-based retrieval-augmented generation framework (EMNLP 2025) from Chao Huang's Data Intelligence Lab (HKUDS): incorporates a graph text-index with dual-level (low/high-level) retrieval to capture entity relationships that flat chunk retrieval misses, with incremental index updates. At 37.8K★ it is among the most-adopted academic RAG stacks anywhere, extended by the multimodal RAG-Anything (22.3K★).

The same lab ships the AutoAgent zero-code agent framework (9.5K★), the AI-Researcher scientific-discovery agent (NeurIPS 2025), and the OpenCity spatio-temporal foundation model — the highest-adoption OSS portfolio at HKU.