claude-mem: persistent memory across agent sessions

Summary

claude-mem is an Apache-2.0 plugin that gives coding agents memory between sessions. Hooks record what the agent does during a session (tool calls and their results), a background worker condenses that into summaries, and the relevant summaries are injected into later sessions. It started as a Claude Code plugin and now also has install paths for Codex, Gemini CLI, Copilot, OpenCode, Cursor, and others. It is very active: v15.4.0 to v15.6.0 shipped on 2026-10-28 and 2026-10-29. Date published is the latest release date.

Key points

• Install in Claude Code: /plugin marketplace add thedotmack/claude-mem, then /plugin install claude-mem, then restart. Or run npx claude-mem install.
• Pitfall: npm install -g claude-mem installs only the library; it does not register the hooks or the worker, so nothing is recorded.
• What it is good for: picking up a project after a break without re-explaining decisions, or recalling how a past bug was fixed. Recent releases add a "My Work" viewer that groups past conversations by focus area and request, plus fixes to how it redacts recorded data.
• Cost: the observer that summarises sessions calls a model; v15.5.0 added opt-in batching for OpenAI-compatible endpoints to reduce that cost.
• Around 105,000 stars (+2,650 in the week to 2026-10-30), with steady daily commits, so the popularity looks organic.

Why it matters

It is the most widely used third-party take on automatic memory for coding agents, as opposed to hand-written files like CLAUDE.md or Claude Code's own auto-memory. Check what it records before using it on sensitive repositories.