# LLMemory MCP server

> Provides a self-hosted knowledge graph memory for AI assistants with hybrid dense and sparse retrieval.

- Repository: https://github.com/djuvinall/LLMemory
- Page: https://hostedmcp.org/servers/djuvinall/llmemory
- Transport: —
- Stars: 0
- License: Apache-2.0

## FAQ

### How does LLMemory handle data retrieval?

It uses hybrid retrieval combining dense vectors from Ollama and sparse vectors in Qdrant, fused with graph expansion and a lexical fallback.

### Does the server use an LLM for writing to the graph?

No, the server does not use a second LLM on the write path; the assistant provides the structure and the server validates it.

### What authentication methods are supported?

The server uses GitHub OAuth with an allowlist system.

### Can I edit the relationships in the memory?

Yes, relationships are first-class records that can be edited, re-typed, or removed.
