m3 Memory
Local-first, MCP-native memory layer for coding agents (Claude Code, Codex, Gemini CLI, Antigravity, OpenCode, OpenClaw) with hybrid search and chat capture.
Tags
Description
pipx install m3-memory, then m3 setup detects installed agents and wires m3 into each (MCP config plus session-start and chat-capture hooks), and m3 doctor checks the install. Stores facts, decisions, preferences and conversation history locally with hybrid search (SQLite FTS5 + vectors + MMR); works fully offline, needs no cloud account or API key, and can optionally use a local LLM (Ollama, LM Studio) to summarize and link memories. The project reports 99.2% LongMemEval-S retrieval at k=8 (self-reported). Python 3.12+; PyPI m3-memory 2026.10.25.0.
Related
- MCP Memory Service: Open-source persistent memory for AI agent pipelines and Claude, with a REST API, knowledge graph and autonomous consolidation.
- Basic Memory: Local-first MCP server that gives AI assistants persistent memory stored as Markdown files in a knowledge base you own.
- agentmemory: Persistent memory server for coding agents (Claude Code, Copilot CLI, Cursor, Gemini CLI, Codex CLI, OpenCode and any MCP client) with hybrid search.
- OmniMemEval: MemTensor's evaluation framework for memory backends (LoCoMo, LongMemEval, BEAM, PersonaMem v2, HaluMem) and memory-equipped agents on AgentBench domains.
Guides & comparisons
- Memory servers for coding agents compared: m3 Memory vs agentmemory vs Basic Memory vs MCP Memory Service (Comparison): Four self-hosted memory servers that give Claude Code, Codex, Cursor and other MCP clients one shared memory across sessions and tools. Covers install, storage, search, how memories get captured, hosted options and license.
Sources
- https://github.com/skynetcmd/m3-memory
- https://github.com/skynetcmd/m3-memory/releases
- https://pypi.org/project/m3-memory/
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)