github.devotd.
NewAI agents

thedotmack/claude-mem v13.39.0 — Persistent Memory Across Agent Sessions

thedotmack/claude-mem is a persistent memory system that compresses records of agent sessions and provides context to later sessions.

Repositorythedotmack/claude-mem ↗

Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

stars
★ 101.3K
Forks
8.9K
Language
TypeScript
License
Apache-2.0
Latest release
v13.41.0
Read original source ↗View on GitHub ↗

thedotmack/claude-mem v13.39.0 is a persistent memory compression system for Claude Code. It collects the work an agent performs and the tool usage during a session, compresses it with AI, and serves the relevant context back to later sessions. The README describes its goal as keeping project context continuous even after a session ends or you reconnect. The latest release is v13.39.0; it is written in TypeScript and licensed under Apache-2.0. Node.js 20 or later is required.

The role of this project is to carry project context across sessions that have ended. It reduces the record of a session into a semantic summary and injects that into the next session. The README covers quick start, how it works, MCP search tools, configuration, and troubleshooting. Alongside Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, and OpenCode are also listed as supported targets. The repository records 101142 stars.

That said, the default memory store is CMEM Pro, a hosted service. The documentation says local observation requires selecting --provider host, so local use should not be treated as the default behavior. The provided information includes no benchmarks comparing summary quality and no results on how much original information survives compression. The structure that links context across sessions is therefore confirmed, but the actual preservation quality is hard to judge from the given information alone.