LightRAG
A Python library that builds a knowledge graph from your documents to answer questions with an LLM.
- Stars
- 40.5k
- Forks
- 5.8k
- Licence
- MIT
- Language
- Python
- Last update
- 7 days ago
- Type
- Library
What it is
LightRAG is a Python library for retrieval-augmented generation, where a language model answers questions using information pulled from your own documents. It extracts entities and relationships from text into a knowledge graph and uses that graph, alongside vector search, to find relevant context for a query. The README also describes a web UI, Docker-based deployment, and support for several storage backends.
- Index a folder of documents and query them through the API or WebUI
- Store the knowledge graph in backends such as Neo4j, PostgreSQL, MongoDB, or OpenSearch, as the README lists
- Parse PDFs and Office files with MinerU or Docling integration described in the news items
- Evaluate retrieval quality with the RAGAS integration mentioned in the README
Great for
- Developers building question-answering over their own document collections
- Teams who want a graph-based alternative to plain vector-search RAG
- Researchers experimenting with graph-based retrieval and evaluation (RAGAS and Langfuse integrations are listed)
- People who want a self-hosted RAG setup with a choice of storage backends
Not ideal for
- Non-programmers who only want a ready-made chat app; the README centers on a Python library and API, though a WebUI is mentioned
- Anyone needing a fully managed hosted service; this is self-hosted open source
- Projects where you cannot pay for or run an LLM and embedding model, since the README relies on external model services
How to get it
- 1
Install with a package manager
You'll need: Python 3.10
Install uv (package manager recommended by the README)
git clone https://github.com/HKUDS/LightRAG.git - 2
Run it
- Clone the repository from GitHub
- Install dependencies following the README's Installation section, which recommends uv
- Configure an LLM and embedding model per the README's configuration instructions
- Run the example or the API/WebUI as described in the README
- 3
Get the most out of it
- Use role-specific LLM configuration to assign separate models to extraction, query, keywords, and VLM tasks
- Choose among the four text chunking strategies (Fix, Recursive, Vector, Paragraph) to suit your documents
- Enable the Reranker, which the README says is the default query mode
- Use the RAGAS and Langfuse integrations to evaluate and trace queries
- Try Docker-based local deployment of embedding, reranking, and storage backends via the setup wizard
Downloads
Strengths
- Built on a published research paper (EMNLP 2025 per the description) with a linked arXiv preprint
- Supports multiple storage backends and a range of configurable chunking strategies
- MIT license and an active project with recent releases and a large star count
Trade-offs
- Setup involves several moving parts (LLM, embedding model, storage), which the README does not simplify for beginners
- Many features are recent and listed in news items, so behavior may change between versions
- The README shown is truncated, so full configuration details were not available to verify
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| LightRAG (this) | 40.5k | Library | intermediate | 7 days ago | Developers building question-answering over their own document collections |
| graphrag | 36.5k | Library | intermediate | 4 days ago | Researchers exploring graph-based retrieval over narrative or private text |
| RAG-Anything | 24.4k | Library | intermediate | 4 days ago | Developers building question-answering systems over PDFs or Office documents with figures and tables |
| llama_index | 52.8k | Library | intermediate | 10 days ago | Python developers building retrieval-augmented generation (RAG) apps over private documents |
| PageIndex | 40k | Library | intermediate | 7 days ago | Developers building question-answering over long PDFs such as annual reports or manuals |
| LEANN | 13.8k | Library | intermediate | 4 days ago | Developers building a private RAG prototype on a personal machine |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.