HippoRAG

HippoRAG is a Python memory and retrieval framework for developers building LLM applications that need to connect facts across documents. It uses a knowledge graph to find related information for questions that require several pieces of evidence.

• Connected retrieval: Find passages through relationships between facts, including questions whose answers span multiple documents.
• Document-grounded answers: Generate answers with a configured LLM using retrieved passages, or return passages with relevance scores and source metadata.
• Persistent knowledge: Add documents to an index that survives restarts, skip duplicate passages, and remove content by text or source identifier.
• Agent access: Share the index with Claude Code, Claude Desktop, Cursor or VS Code through MCP; the Claude Code plugin provides a folder-indexing command and an index shared across projects.

HippoRAG is open source under the MIT license and runs on your own hardware or server. It supports local vLLM models such as Llama-3.3-70B-Instruct, with embeddings from NV-Embed-v2 or GritLM. Local GPU deployment needs enough memory for both the LLM and embedding model.

You can also use OpenAI, Azure OpenAI, Amazon Bedrock or OpenAI-compatible hosted endpoints. Those services receive the requests sent to them and require their own credentials. Embeddings stay in local Parquet files by default, with Qdrant, ChromaDB and Milvus as alternatives. Existing indexes are tied to their model configuration; changing the LLM or embedding model requires a separate index.