Claude-Mem 🧩 Skill Freemium

What it does

Claude-Mem, by Alex Newman, gives Claude Code a long-term memory that lasts across sessions. It hooks into the session lifecycle, records what the agent does with its tools, and uses a model to compress those records into short observations and summaries. It then injects the relevant parts into later sessions. The result is that a new session starts with context about decisions, bug fixes and project history instead of from nothing.

What is inside

• Lifecycle hooks: session start, prompt submit, post-tool-use, stop and session end.
• A local worker service: runs on Bun, with an HTTP API and a web viewer that shows the memory stream.
• Storage: SQLite with full-text search, plus a Chroma vector database for semantic search.
• Search tools: search, timeline and get_observations, served over MCP, plus a mem-search skill. Together they fetch a compact index first and full details only for the records you need.
• Controls: <private> tags to keep content out of storage, context-injection settings, and language modes.

Works with

Claude Code is the main target, installed as a plugin. The installer also supports OpenCode, Antigravity CLI, OpenClaw gateways and a Grok bot. The repository description claims support for more agents, but the README documents only these.

How to install

npx claude-mem install

Or, inside Claude Code:

/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem

Maintenance and safety

Apache 2.0 and very actively developed, with stable and early-access branches. By default the installer asks you to sign in to a hosted "CMEM Pro" memory service, free for 30 days. You can skip this with an explicit --provider flag or CLAUDE_MEM_ONLINE_OPTIN=false. The plugin stores detailed records of your sessions on disk, and a cloud sync option is available, so decide what should leave your machine. It also auto-installs Bun and uv if they are missing. The README promotes a third-party crypto token linked to the project. That token has nothing to do with using the tool.

Who should use it

Heavy Claude Code users on long-running projects who lose time re-explaining context. Teams with strict data rules should run it local-only.