claude-mem: The 115,000-Star Tool That Gives Claude Code a Memory Across Sessions
Every coding session with Claude Code starts from zero. The agent re-reads the same files, re-asks the same questions and walks back into mistakes you already fixed — unless something remembers for it. That something is claude-mem, an open-source tool with 115,678 stars on GitHub that gives coding agents persistent memory across sessions. There is a privacy caveat to read before installing it, and this article covers it honestly, because a memory tool is only useful if you can trust what it does with your work.
claude-mem is a persistent memory layer for AI coding agents: a second agent beside your coding agent takes notes while it works, compresses that activity into summaries with AI, and injects the relevant context back into future sessions, so the next session opens with a short index of what matters instead of a blank slate.
The problem it solves
The pain is familiar to anyone who uses an AI coding assistant daily. A session ends, the context is gone, and the next session has no idea what was decided, which bug was fixed, or why an approach was rejected. Developers end up repeating themselves and re-debugging solved problems. The project's README states it plainly: every new session starts from zero.
That problem grows as coding agents take on longer work: Anthropic's Claude Duet agent family is built for multi-session projects, and memory is what makes that practical. Pricing matters too — the Claude Haiku 5.5 price cut made heavy Claude Code use cheaper, so more sessions means more value in not re-reading the same files.
How claude-mem works
The mechanism has three steps. First, hooks in the coding agent hand each tool call — what it reads, edits and runs — to a small local worker on the developer's machine. Second, an observer agent turns that activity into short observations and writes a summary when the session ends. Third, the next session opens with a compact index of past observations, layered so the agent opens full details only when it needs them, keeping token costs down.
Memory is searchable, not just injected. A mem-search skill lets the agent search project history in three steps — index, timeline, details — so it spends tokens only on what it opens. A My Work dashboard shows live requests and a saved review queue.
The numbers: why this repo is trending
That growth is why claude-mem leads this week's GitHub round-up. The repository, thedotmack/claude-mem, was created on 31 August 2025 and pushed today, 10 November 2026, with release v15.15.1 the same day. It topped GitHub's daily Trending page, gaining 1,526 stars in a single day around 2 November, and an AI-radar report on 6 November recorded 113.8k stars — the count has climbed past 115,000 since.
| Figure | Value (verified live, 10 Nov 2026) |
|---|---|
| GitHub stars | 115,678 |
| Forks | 10,040 |
| Created | 31 Aug 2025 |
| Last push | 10 Nov 2026, 10:44 UTC |
| Licence | Apache-2.0 |
Those numbers put it ahead of the biggest developer tools we have covered in this series: the Impeccable GitHub skill, the hub post for this cluster, covered a tool with around 96,000 stars; the Hallmark GitHub skill sat near 41,000; and the claude-seo Claude Code SEO plugin had just under 25,000. Star counts are a popularity signal, not a quality certificate — but a tool gaining more than a thousand stars in a day is one developers are actively choosing.
What it works with, and how to install it
claude-mem is not limited to Claude Code. Its README says it works with Claude Code, Codex, Cursor, OpenCode and 11 more agents, including OpenClaw, Gemini, Hermes and Copilot. Copilot CLI, Goose, Roo Code and Warp can search memory, but their sessions are not recorded.
Installation is a single command, taken from the project's own README:
npx claude-mem install
The installer finds the agents on the machine and asks which ones to set up, then observations start streaming from the first prompt. A single agent can be set up with an --ide flag, for example npx claude-mem install --ide opencode. Claude Code users can also install from the plugin marketplace with /plugin marketplace add thedotmack/claude-mem followed by /plugin install claude-mem, then restart Claude Code. The README warns that npm install -g claude-mem installs the SDK only and does not register the plugin hooks.
There is a commercial layer to be aware of. Setup asks the user to sign in through the browser with an email magic link, and choosing the claude-mem observer starts a 30-day free trial on a checkout page that asks for a payment method. When the trial ends, memory falls back to the user's Anthropic plan unless they subscribe. Users who prefer to skip the sign-in can pass an explicit --provider flag. The provider can also be the user's own OpenRouter or Gemini key. Anonymous usage analytics are on by default and can be switched off with npx claude-mem telemetry disable.
The privacy caveat: read this before you install
On 29 October 2026, a user filed GitHub issue #5176, "Installing claude-mem uploads your last 30 days of prompts to cmem.ai, no opt-in". The report said the installer opened a browser setup page that read local chat histories and uploaded the full text of prompts typed in the previous 30 days, with session IDs and project names, starting as soon as the setup page loaded, with no way to stop it. The only checkbox was ticked by default and disabled during the upload, and the project confirmed unticking it did not stop the upload.
Release v15.8.3, published on 2 November 2026, fixed the installer: browser setup no longer uploads prompts without asking. The scan now reads local history only, then stops and asks, and nothing leaves the machine until the user answers. A consent panel lists what is sent per tool; only ticked tools are uploaded, and choosing "Keep everything local" uploads nothing.
Two things remain unresolved, both re-verified at drafting time. First, issue #5176 is still open — checked on 10 November 2026, it was last updated on 2 November and has not been closed. Second, the v15.8.3 notes state plainly that prompts already uploaded by earlier versions are not removed by the update. The fix stops future uploads without consent, but it does not undo uploads that already happened.
| Privacy point | Status at drafting (10 Nov 2026) |
|---|---|
| Installer uploads prompts without asking | Fixed in v15.8.3 (2 Nov 2026) — setup now asks first |
| GitHub issue #5176 | Still open; last updated 2 Nov 2026 |
| Already-uploaded prompts | NOT removed by the update, per the v15.8.3 notes |
| Memory storage | Local in ~/.claude-mem; cloud sync opt-in |
| Anonymous usage analytics | On by default; disable via telemetry disable |
| Keeping things out | Wrap in <private> tags — never stored |
Who should use it, and who should wait
For solo developers and small teams who live in Claude Code or Codex, the value is straightforward: fewer repeated explanations, faster restarts, and a searchable record of decisions and bug fixes. The 30-day observer trial makes the hosted memory easy to try, and local-first storage plus the <private> tag gives users control over what gets remembered.
Anyone working with client code or sensitive data should go in with eyes open. The installer now asks before uploading, but the open issue and the unremoved past uploads mean trust has to be rebuilt rather than assumed. Install with a local provider, keep cloud sync off, disable analytics, and use <private> tags on sensitive material.
Key Takeaways
- claude-mem gives Claude Code and other coding agents persistent memory across sessions: it records tool use, summarises it, and injects relevant context into future sessions.
- The repo has 115,678 stars and 10,040 forks, verified live on 10 Nov 2026; created 31 Aug 2025, Apache-2.0 licensed, pushed today with release v15.15.1.
- It works with Claude Code, Codex, Cursor, OpenCode and 11 more agents; install with
npx claude-mem installor from the Claude Code plugin marketplace. - Privacy caveat: an earlier installer uploaded 30 days of prompts without asking. v15.8.3 (2 Nov) fixed this so setup asks first, but issue #5176 is still open and already-uploaded prompts are not removed.
- Memory lives locally in
~/.claude-memwith opt-in cloud sync; wrap sensitive material in<private>tags and disable analytics withnpx claude-mem telemetry disable.
Conclusion
claude-mem earns its 115,678 stars by fixing the most annoying thing about AI coding assistants: the daily amnesia. It is actively maintained, openly licensed, and open about its trial and privacy fix. But the October installer incident is a genuine black mark, and the still-open issue plus the unremoved past uploads mean the project has not fully closed the book on it. Install the current version, keep your memory local, and let the tool remember your code — on your terms.
Comments
Post a Comment