LightRAG

// Overview

LightRAG is an open-source Python framework for retrieval-augmented generation (RAG) from HKUDS. Instead of only chunking text and matching embeddings, it extracts entities and the relationships between them to build a knowledge graph, then uses that graph alongside vector search when answering a query. The package is published on PyPI as lightrag-hku.

It is aimed at developers who are building question-answering or assistant features over their own documents and want retrieval that can reason across entities and documents, not just return the nearest text chunks. Because the graph supports incremental updates and document deletion with automatic regeneration, you can keep a corpus current without rebuilding everything from scratch.

Within the RAG framework space, LightRAG sits next to tools like vector-only pipelines and other graph-RAG systems. It offers several query modes (naive, local, global, hybrid, and mix), pluggable storage backends such as PostgreSQL, MongoDB, Neo4j, and OpenSearch, a built-in WebUI, and an optional API server, so you can start small and grow into a larger deployment.

// What it does

• Builds an entity-and-relationship knowledge graph from your documents to support retrieval that reasons across entities and sources
• Five query modes — naive, local, global, hybrid, and mix — so you can trade off precise local matching against broad cross-document reasoning
• Incremental updates and document deletion with automatic knowledge-graph regeneration to keep retrieval accurate over time
• Pluggable storage backends including PostgreSQL, MongoDB, Neo4j, and OpenSearch for graph, vector, and key-value data
• Works with OpenAI and open-source LLMs (for example Qwen3), plus reranker support to improve mixed queries
• Ships a WebUI for inserting, querying, and visualizing the graph, and an optional API server via the [api] extra

// Getting started

Install the package from PyPI, set your model API key, then initialize storage, insert documents, and query. LightRAG's official entry point is the OpenAI demo script.

pip install lightrag-hku
# with API server + WebUI extras:
pip install "lightrag-hku[api]"

export OPENAI_API_KEY="sk-...your_openai_key..."
curl https://raw.githubusercontent.com/gusye1234/nano-graphrag/main/tests/mock_data.txt > ./book.txt

import asyncio
from lightrag import LightRAG, QueryParam
from lightrag.kg.shared_storage import initialize_pipeline_status

async def main():
    rag = LightRAG(working_dir="./rag_storage")
    await rag.initialize_storages()
    await initialize_pipeline_status()

    with open("./book.txt") as f:
        await rag.ainsert(f.read())

    answer = await rag.aquery(
        "What are the top themes in this story?",
        param=QueryParam(mode="mix"),
    )
    print(answer)

asyncio.run(main())

python examples/lightrag_openai_demo.py

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Build a question-answering assistant over an internal document set that needs to reason across entities, not just match the nearest text chunk
• Keep a knowledge base current with incremental document inserts and deletions without rebuilding the whole index
• Answer broad, cross-document questions using global or mix query modes where simple vector retrieval falls short
• Stand up a graph-backed RAG service using PostgreSQL, MongoDB, Neo4j, or OpenSearch and the optional API server

// How LightRAG compares

LightRAG alongside other open-source rag frameworks & platforms tools AI/TLDR tracks, ranked by GitHub stars.