Is claude-mem worth installing? If you work in Claude Code on the same long-lived project every day, and every new session makes you re-explain the background and re-hit the same pitfalls, then yes; if you only write the occasional small script, it adds a layer of maintenance you do not need. It is an open-source project on GitHub that gives coding agents long-term memory: it automatically records tool calls and decisions from each session, then injects the relevant context back into the next one.
What it is, and why it is so popular
Repository facts: platform GitHub, author and organization thedotmack, project name claude-mem, Apache-2.0 open-source licence. In week 45 of November 2026 it reached the GitHub Trending weekly chart, gaining about 9,318 stars in a single week, and stood at roughly 117,000 stars as of November 17. The reason for the buzz is straightforward: every new Claude Code session starts from zero, while claude-mem uses hooks to record tool calls, a second observer model compresses the process into observation notes, and stores them in a local SQLite database plus a vector index — the next session starts with a compact index and fetches details only when needed.
Deployment cost and getting started: up and running in half an hour?
Installation itself is easy: with Node.js 20 or later installed, one command — npx claude-mem install — does it, and Bun and uv are filled in automatically; you can also install it from the plugin marketplace inside Claude Code. It supports Claude Code, Codex, Gemini, OpenCode, OpenClaw and more hosts. The real cost comes later: compressing memory calls a model, and the default route is a hosted observer service with a 30-day free trial, then a subscription — or you switch to your own OpenRouter or Gemini key, or your Anthropic plan. Data lives on your machine by default; cloud sync is optional.
Three pitfalls to think through before connecting it
First, memory swells: the more it records, the messier the injected context gets — regular cleanup and mode settings are not optional. Second, the privacy boundary: observation notes can sweep up keys and sensitive information, so exclude them with private markers, and note that the hosted service sends observations to the cloud. Third, the stability of the observer service: compressing memory calls a model, so when rate limits or network hiccups hit, pending observations pile up and slow session startup — if something goes wrong, start with the built-in doctor command. Compared with Mem0, covered on this site, claude-mem sits closer to the coding workflow: it records tool calls rather than conversation summaries, while Mem0 is more of a general memory layer suited to reuse across applications. If you just want an agent to remember project decisions, pick claude-mem; if you need one memory service spanning several applications, look at Mem0.