LightRAG
GraphRAG without the GraphRAG bill
What it is
A graph-augmented RAG framework from HKU's data-science lab: extracts entity-relationship knowledge graphs from documents and combines dual-level retrieval with vector search. Ships as a Python library plus a server with an interactive graph visualizer, supporting many storage backends and LLM providers.
Why it's interesting
The practical GraphRAG — Microsoft-style relational reasoning at a fraction of the indexing cost, under MIT, with incremental graph updates instead of full re-indexing. One of the fastest-growing RAG projects of the past two years, from the same lab as Vibe-Trading.
Use cases
- Q&A over interconnected document collections
- Building queryable knowledge graphs from text
- Multi-hop questions plain vector RAG fails at
Who it's for
RAG builders, knowledge-management engineers, researchers
Setup
Easy. pip install lightrag-hku or Docker; an LLM (32B+ recommended for extraction quality) and embedding model
Limitations & cautions
Graph quality depends on the extraction LLM and adds token cost at index time; the codebase moves fast with occasional breaking changes.
Editorial takeaway
The version of graph-RAG you can actually afford to run on your whole corpus — which is the only version that matters.