GraphRAG from Microsoft Research extracts entities and relations into a knowledge graph before retrieval, dramatically improving answers to global, multi-hop questions over large corpora.
| Category | LLM / RAG framework |
| Type | RAG pipeline |
| License | MIT |
| Runs locally | Partial |
| Built with | Python |
| Skill level | Advanced |
| Best for | complex question-answering over big document sets |
Other open-source llm / rag framework tools worth comparing:
LangChainCompose chains, tools and agents
LlamaIndexThe data framework for RAG
HaystackProduction pipelines for search and RAG
DSPyProgram — not prompt — language models
txtaiAll-in-one embeddings database
RAGFlowDeep-document-understanding RAG
Semantic KernelMicrosoft's enterprise agent framework
LiteLLMOne API for 100+ LLM providers
InstructorReliable structured outputs from LLMs
OutlinesGuarantee valid JSON and grammars
GuidanceInterleave control and generation
LLMWareEnterprise RAG with small specialised models
FlashRankUltra-light reranking for better RAG
Sentence TransformersThe standard way to make embeddings
RagasMeasure whether your RAG is any good
PromptfooUnit tests for your prompts
LangfuseSee what your LLM app actually did
PhoenixTrace, evaluate and debug LLM appsGraphRAG is free and open-source (MIT license), so you can use, self-host and modify it at no cost.
GraphRAG can be used locally or in the cloud depending on your setup.
Popular open-source alternatives include LangChain, LlamaIndex, Haystack. See the comparisons above to choose.
Browse the full directory of open-source AI tools, models and projects — updated daily.
Browse all tools →