Claude Mem Review

claude-mem gives coding agents memory between sessions. Hooks in your agent pass each tool call (files read, edits, commands) to a small local worker; a second "observer" model turns that activity into short observations and writes a summary when the session ends. The next session in the same project starts with a compact index of those observations, and the agent pulls full details only when it needs them through three MCP tools (search, timeline, get_observations). Memory lives in ~/.claude-mem as a SQLite database, with Chroma vector search on top, and a local web viewer shows sessions, observations and settings. Anything wrapped in <private> tags is never stored.

Install with npx claude-mem install, which detects the agents on your machine, or from the Claude Code plugin marketplace (/plugin marketplace add thedotmack/claude-mem). The README lists support for Claude Code, Codex, Cursor, OpenCode, T3 Code, Antigravity CLI, OpenClaw gateways and others; Copilot CLI, Goose, Roo Code and Warp can search memory but their sessions are not recorded. The observer can run on the hosted CMEM service, your own OpenRouter or Gemini key, or your Anthropic plan. Anonymous usage analytics are on by default and can be turned off with npx claude-mem telemetry disable.

The pricing page does not publish numeric allowances for Pro, Heavy or Max; included usage is shown at checkout. The free engine itself costs nothing, but the observer has to run on some model: either the paid hosted observer, your own OpenRouter or Gemini key, or your Anthropic plan (which uses your plan's usage). The CMEM Pro trial needs a credit or debit card (prepaid cards are not accepted), and billing can start when the trial ends or when included usage runs out, whichever comes first. If the trial ends without a subscription, memory falls back to your Anthropic plan.

Prices checked on the vendor's pricing page on November 10, 2026. See CMEM pricing.

What the claude-mem README and site list:

claude-mem suits developers who work in the same repositories across many agent sessions and are tired of re-explaining decisions, fixes and project structure. It fits best if you want memory stored on your own machine and you use more than one coding agent, since one memory store serves several of them. The free engine is enough if you already pay for an Anthropic plan or a model API key; CMEM Pro makes sense if you want memory generation off your agent's plan and synced across computers.

For short one-off tasks it adds little, and Claude Code's own CLAUDE.md and auto memory may be enough. If you are building memory into your own AI product rather than a coding agent, look at Mem0 or Supermemory. For another open-source coding-agent memory layer, compare agentmemory and MemPalace.

Coverage is mostly guides and desk reviews rather than measured tests. Implicator ranked claude-mem first of five coding-agent memory plugins on setup, local storage and cross-agent support, while stating it was "a desk review, not a lab test"; it notes that compression sends observations to a model provider you pay for, that a rival had better-documented recall scores, and that claude-mem becomes a second memory writer beside Claude Code's built-in auto memory. DataCamp's guide walks through installation and the search workflow. Some reviewers suggest it is overkill for quick single-session tasks. No independent benchmark of its token savings was found. Beware of unofficial forks on GitHub that carry cryptocurrency links; the official repository is thedotmack/claude-mem.