RAG et connaissances
graphrag
Microsoft's graph-based RAG: index a corpus into a knowledge graph, then query it with an LLM
★ 36,3 k étoilesMIT
Voir le guide →nano-graphrag est un projet de RAG et connaissances open source, créé par gusye1234 (gusye1234/nano-graphrag sur GitHub). Il compte 3 993 étoiles sur GitHub et 432 forks, il est publié sous la licence MIT et a été mis à jour pour la dernière fois sur GitHub le 27 janvier 2026.
nano-graphrag: a small, hackable GraphRAG implementation in about 1,100 lines — no Microsoft GraphRAG bulk
Export your OpenAI key (or configure Azure, Bedrock, or a local Ollama setup from the examples).
Install from PyPI, or clone and install in editable mode to hack on the core.
Create a GraphRAG instance with a working directory and insert your documents.
Ask questions in local, global, and naive modes to compare graph-aware versus plain retrieval.
Roughly 1,100 lines implement the full GraphRAG pipeline — built to be hacked.
Entity-centric (local), community/theme-centric (global), and plain vector (naive) answering.
New text is deduplicated by md5 hashing and merged into the existing graph.
ainsert and aquery mirror the sync API for async applications.
Swap chunking, LLM, and embedding functions and prompts; choose nano-vectordb, faiss, hnswlib, or milvus-lite, and networkx or Neo4j storage.
OpenAI by default, with Azure OpenAI, Amazon Bedrock, and Ollama examples.
Clone the project and enter its directory (or install from PyPI in the next step).
git clone https://github.com/gusye1234/nano-graphrag.git
cd nano-graphragInstall the package and export your OpenAI API key.
pip install nano-graphrag
export OPENAI_API_KEY="your-key"Index a text file and ask a question in local mode, following the README quickstart.
from nano_graphrag import GraphRAG, QueryParam
graph_func = GraphRAG(working_dir="./dickens")
graph_func.insert(open("book.txt").read())
print(graph_func.query("What are the top themes?", param=QueryParam(mode="local")))Developer gusye1234 maintains nano-graphrag as an independent, community project inspired by Microsoft's GraphRAG.
Yes, MIT-licensed. Model calls bill to your own OpenAI/Azure/Bedrock keys, or run locally via Ollama examples.
By default it expects an OpenAI API key; the repo includes examples for Azure OpenAI, Bedrock, and local Ollama models if you prefer.
Microsoft's is the full research platform with extensive configuration. nano-graphrag trades breadth for a ~1,100-line core you can fully read and modify, with the same local/global query idea.
Run `pip install nano-graphrag`, set OPENAI_API_KEY, create GraphRAG(working_dir=...), insert text, and query with QueryParam(mode="local").
Microsoft's graph-based RAG: index a corpus into a knowledge graph, then query it with an LLM
Unified framework for enterprise RAG pipelines with small, specialized models - local, private and secure.
Federated AI search and RAG across 100+ apps — answers with sources, without moving your data
Lightweight OCR toolkit turning PDFs and images into structured, LLM-ready data in 100+ languages
Self-hosted rag & knowledge: cometchat docs search + implementation bundles: add cha.
Open-source AI workspace for projects, wikis, and team docs.
Les autres projets de la catégorie RAG et connaissances avec le plus d'étoiles du catalogue LibreAI — la sélection réaliste d'alternatives si nano-graphrag ne vous convient pas.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or…
RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context…
Lightweight OCR toolkit turning PDFs and images into structured, LLM-ready data in 100+ languages
MinerU — turns PDFs and Office documents into LLM-ready Markdown and JSON for agents and RAG pipelines, Apache-2.0 plus terms
Stop renting your intelligence. Own it with AnythingLLM. Everything you need for a powerful local-first agent experience
Projets de la catégorie RAG et connaissances proches dans le catalogue, les plus proches en étoiles de nano-graphrag — continuez à explorer de part et d'autre.
Ragent (容纳智能体): a production-grade Java Agentic RAG platform with hybrid retrieval and an admin console
The open-source, permission-aware context layer that lets agents search all your company knowledge.
RAGApp: deploy a RAG app over your documents with Docker — configure it like a custom GPT
Knowhere turns documents into structured, hierarchy-native chunks ready for agents and RAG.
AI-native database for hybrid search across dense, sparse, and tensor vectors plus full text
A personal knowledge base that builds itself — drop in sources, get a persistent interlinked wiki.
Gratuit et ouvert : récupérez le code source, mettez une étoile au dépôt et commencez à créer.