yoloshii/ClawMem
View on GitHub ↗On-device memory layer for AI agents. Claude Code, Hermes and OpenClaw. Hooks + MCP server + hybrid RAG search.
200 ★32 forksTypeScriptUpdated 29d ago
What you need to know
On-device memory layer with retrieval-augmented search for Claude Code, OpenClaw and Hermes agents - no API keys or cloud dependencies
Install
npm install -g clawmem clawmem bootstrap ~/notes --name notes
Usage
- •clawmem setup hooks && clawmem setup mcp to integrate with Claude Code
- •clawmem doctor to verify installation
Key features
- ✓Hybrid retrieval combining BM25, vector search, RRF, query expansion and cross-encoder reranking
- ✓Hooks that surface relevant context on every prompt and generate handoffs at session end
- ✓MCP server with 33 tools plus optional HTTP REST API
- ✓Native OpenClaw and Hermes plugin integrations sharing the same SQLite vault
- ✓Multi-vault support for separate memory domains
- ✓Pin/snooze lifecycle management and contradiction detection
Best for
Teams wanting a single local memory vault shared across multiple AI coding agents
Caveats
- ⚠Windows native is not recommended - use WSL2
- ⚠Requires Bun 1.0+
- ⚠Optional upgrades may need llama.cpp GPU servers or cloud embedding
Platforms: Linux · macOS · Windows (WSL2)Clients: Claude Code · OpenClaw · Hermes Agent · Any MCP client
Reviewed 2026-08-11
Topics
ai-agent-memoryai-agentsbunclaude-codeembeddingshermeshybrid-searchllama-cpplocal-firstmcp-servermcp-toolsmemorymodel-context-protocolon-device-aipluginragretrieval-augmented-generationsqlitetypescriptvector-search
- Stars
- 200★
- Forks
- 32
- Language
- TypeScript
- License
- MIT
- Created
- 2026-02-06
- Last push
- 2026-08-06