claude-mem gives Claude Code and other agents persistent cross-session context
Decision Brief
claude-mem captures tool-use observations via lifecycle hooks such as SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop and SessionEnd, writes them to SQLite, generates semantic summaries, and automatically injects relevant context into new sessions. Four MCP tools implement a three-tier workflow: search returns a compact index with IDs (~50-100 tokens each), timeline provides temporal context, and get_observations fetches full detail only for filtered IDs (~500-1000 tokens each); the project claims roughly 10x token savings versus pulling full text directly. Vector retrieval is backed by Chroma, with a local Worker managed by Bun and a web viewer. Installation supports Claude Code, OpenCode 1.3.4+, T3 Code, Antigravity CLI, OMP, Pi, DeepSeek Harness and the OpenClaw gateway. Running npx claude-mem install triggers a browser email login, after which you get a 30-day free trial and default to the hosted CMEM Pro; you can also pass --provider to use your own OpenRouter or Gemini key, or set CLAUDE_MEM_ONLINEOPTIN=false to skip login. Teams using Claude Code can install directly from the plugin marketplace with /plugin marketplace add thedotmack/claude-mem; note that npm install -g claude-mem only installs the SDK and does not register hooks or start the Worker service. Licensed under Apache-2.0, making it easy to embed in internal enterprise toolchains.
Sources
- Skill Radar(GitHub 趋势)
Trending hands-on MCP servers, agent skills, and AI-coding tools discovered from GitHub search momentum.
- GitHub:thedotmack/claude-mem
留言
登入后即可留言,和其他 builder 交换实测心得。
还没有留言,抢头香。