Claude Mem Pricing: Free Open Source vs Pro, Heavy, and Max Plans
Claude Mem's engine is free and open source, but its hosted memory tiers run $30 to $100 a month. Here's how the pricing actually breaks down.
What does Claude Mem actually cost?
Claude Mem itself is free. It’s an open source project under Apache 2.0, and the core engine, the one that watches your coding agent’s sessions and writes memory to a local database, doesn’t cost anything to run. What costs money is the model that reads and compresses your sessions into memory, called the observer, and how you choose to pay for it determines your total bill. You can run it on your own Anthropic plan at no extra charge beyond your existing usage, or pay for a hosted plan (CMEM Pro) starting at $30 a month after a 30-day free trial, with Heavy at $70 and Max at $100.
TL;DR
- The software is free and open source, with more than 112,000 GitHub stars and no license fee for the core memory engine itself.
- The real cost comes from the observer model, a background process (Haiku, by default, on Anthropic) that compresses tool calls and session activity into stored memory notes.
- Running it on your own Anthropic plan means memory usage shares your existing rate limits, so there’s no separate bill, but you’re spending tokens you’d otherwise use for coding.
- The hosted option, CMEM Pro, starts at $30 a month after a 30-day trial, with Heavy at $70 and Max at $100 for heavier usage.
- A
--providerflag at install time skips the hosted sign-up entirely, letting you point memory at your own Anthropic, OpenRouter, Gemini, or any OpenAI-compatible key. - There’s at least one reported case of memory background tasks eating into a user’s quota unexpectedly, which is worth watching if you’re cost-sensitive.
- For occasional, one-off coding tasks, the free tier (your own API key, no hosted plan) is probably enough, and a well-written project config file may make memory unnecessary altogether.
How does Claude Mem’s pricing structure work?
Claude Mem separates the tool from the model that powers it. The tool, the install script, the hooks into Claude Code, Codex, Cursor and other agents, the local SQLite and vector database, is open source and free under Apache 2.0. Anyone can clone it, audit it, or self-host the whole pipeline.
What isn’t free is the “observer,” the model that watches raw tool calls (file reads, edits, commands) and turns them into structured notes with a title, type, narrative, and list of touched files. That’s a model doing real inference work every time your coding agent does something. Someone has to pay for that inference, and Claude Mem gives you two ways to handle it.
The first is to point it at your own account. Passing --provider during install skips the hosted sign-up and routes memory requests through your own Anthropic plan, or through OpenRouter, Gemini, or any OpenAI-compatible endpoint if you’d rather use a different model as the observer. In this mode, there’s no separate subscription. Memory tasks just draw from whatever plan or API budget you already have.
The second is the hosted option, called CMEM Pro, which is pre-selected by default during installation. It comes with a 30-day free trial, after which it converts to a paid subscription: $30 a month for the base Pro tier, $70 a month for Heavy, and $100 a month for Max. These tiers presumably scale with how much observation and search volume you generate, though the exact rate limits or quotas per tier aren’t published in detail anywhere easy to find, so it’s worth checking current usage caps before committing to a tier.
Is the free, self-hosted option actually free?
Mostly, yes, but “free” needs a caveat. If you route Claude Mem through your own Anthropic plan instead of paying for CMEM Pro, you’re not handed a separate invoice. But the observer model still runs, and it still consumes tokens and, depending on your plan, counts against the same rate limits you use for actual coding work. Claude Mem’s own documentation acknowledges that the token savings from its “progressive disclosure” approach (feeding the agent a lightweight index instead of dumping full history into context) depend heavily on your workload and provider. That’s an honest caveat, and it cuts both ways: it could save you a meaningful number of tokens over re-explaining context every session, or it could add background overhead you don’t fully see.
There’s also a reported case worth flagging: a user on the Claude provider found around 230 hidden background runs in four days coming from the “My Work” dashboard refresh feature, which ate into their quota without an obvious trigger. That’s not necessarily representative, but it’s a reminder that “free” in this context means “billed through your existing plan,” not “costs nothing.”
What do the Pro, Heavy, and Max tiers get you?
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
Based on what’s publicly stated, CMEM Pro’s three paid tiers are priced at $30, $70, and $100 a month, with Heavy and Max presumably sized for people running more sessions, more projects, or more agents in parallel. The appeal of paying for hosted memory is that you don’t have to manage the observer’s token usage against your own Anthropic account at all. It’s billed separately and presumably comes with its own capacity, rather than quietly consuming your coding budget.
If you’re running multiple agents across several codebases, or using Claude Mem across tools beyond Claude Code (it also supports Codex, Cursor, OpenCode, Windsurf, GitHub Copilot, and several others), a flat hosted plan might be simpler to reason about than watching memory overhead eat into whatever plan you use for actual development.
For solo developers working on one or two projects, the cost of running the observer through your own existing plan is likely to be the cheaper and simpler path, assuming your plan has headroom.
Is Claude Mem worth paying for?
That depends entirely on your workflow. If you work on the same codebase over weeks and keep re-explaining the same architectural decisions or bug fixes to a fresh session every morning, the value proposition is real: Claude Mem’s index approach means a new session gets a compact summary (reportedly a few hundred tokens for a handful of past observations) instead of nothing, and the agent can pull specific past context only when it needs to.
If your usage is mostly short, one-off tasks, the overhead of running an observer model, hosted or self-funded, probably isn’t worth it. A detailed project configuration file (the kind of persistent instructions file many coding agents already support) covers a lot of the same ground for free, since it doesn’t require a second model running in the background.
Anyone considering it should also weigh a documented privacy issue: an earlier version (before 15.8.3) uploaded recent prompt history to Claude Mem’s cloud service by default during a browser setup step, before asking permission, which ran against what the privacy policy stated at the time. That’s been fixed in current releases, but it’s a separate consideration from pricing and worth understanding before installing any version.
Frequently Asked Questions
Is Claude Mem completely free to use?
The software itself is free and open source. What isn’t necessarily free is the observer model that compresses your coding sessions into memory, since that requires inference. You can route that through your own Anthropic or other API plan at no extra subscription cost, or pay for the hosted CMEM Pro service.
How much does CMEM Pro cost after the free trial?
CMEM Pro offers a 30-day free trial, after which pricing starts at $30 a month for the base tier, $70 a month for Heavy, and $100 a month for Max.
Can I avoid the hosted plan entirely?
Yes. Passing the --provider flag during installation skips the hosted sign-up and lets Claude Mem run on your own Anthropic account, OpenRouter key, Gemini key, or any OpenAI-compatible endpoint instead.
Does using my own Anthropic plan mean memory is free?
It means there’s no separate bill, but the observer model’s token usage still draws from your plan’s existing limits. Depending on your workload, that can add noticeable overhead to your regular usage.
Which plan should I pick if I’m cost-conscious?
Nothing gets lost.
If you’re only doing occasional or short coding tasks, running Claude Mem through your own existing API plan, without a hosted subscription, is likely sufficient and avoids extra monthly cost. Heavier, multi-project or multi-agent users may find a flat hosted tier simpler to manage.
