Claude Memory is a feature that remembers things from your past chats and brings them into new conversations, so you don't have to explain yourself every single time. On Free, Pro, and Max it is on by default according to Anthropic, so there is nothing to activate. On Team and Enterprise it is the other way round: Memory is off until an owner turns it on.
I use Memory for several of my AI employees built on Claude, including Sonja, my assistant who drafts emails in my voice. This post covers what Memory actually is, how to check, pause, or reset it, and where it stops and CLAUDE.md and Claude Projects begin, all sourced directly from Anthropic. I collect the other pieces of my Claude setup on the Claude overview page.
What Claude Memory is, and what it isn't
Memory gathers knowledge from your chats automatically in the background and makes it available in future conversations, according to Anthropic's help article on chat search and memory (support.claude.com). That sets Memory apart from two other building blocks that sound similar but work differently.
A Claude Project is a fixed workspace with instructions and files you set yourself, covered in more detail in How to Use Claude Projects Right. CLAUDE.md, in turn, is a file you write deliberately and by hand in Claude Code, described in CLAUDE.md: Claude Code's Memory. Memory, by contrast, builds itself from what happens in your chats without any deliberate action on your part, and it's tied to your claude.ai account, not to Claude Code.
How to check your Memory setting
A lot of people search for how to activate Claude Memory. The honest answer: on Free, Pro, and Max, on the web, in Claude Desktop and in the mobile app, Memory is already on according to Anthropic, so there is nothing to switch on. What is worth doing instead is looking at what is actually set:
- Open your Claude account settings.
- Go to the Memory section.
- Check whether Claude is drawing knowledge from chats, and look at what it has stored so far.
On Team and Enterprise you will find nothing to toggle there until an owner has enabled the feature. Chat search, which lets you find past conversations, is available only on paid plans according to the same source, not on the Free plan.
Pause or reset: the difference
If Memory records more than you want, Anthropic gives you two different routes, and they do not do the same thing. Pausing holds the memory where it is: existing entries stay, nothing new gets added. Resetting deletes every stored memory permanently and cannot be undone.
For day-to-day work, pausing is almost always the right call, for instance while you spend a few weeks on a mandate that has nothing to do with the rest of your work. Resetting is worth it when so much wrong material has piled up that a clean start is faster than correcting entries one by one.
What Claude remembers, and how to review, correct, and delete it
Anthropic describes Memory as something you can review, edit, and delete at any time, not as an invisible background process. In practice that means you can correct details it got wrong, the same way you'd fix a mistake in a note, and remove individual entries or the whole memory if you want a clean start.
Anthropic itself calls import and export of Memory experimental, according to the help article on the topic (support.claude.com). If you want to bring Memory over from another system or keep an export as a backup, treat it as a bonus feature rather than a reliable default path.
What Memory is good for, and what it isn't
Memory is strong at the things that recur across many chats without you ever writing them down as a rule: a preferred tone, the way you round numbers, a liking for short answers over long ones. You rarely notice patterns like that until you correct them for the tenth time, and that is exactly where a memory that learns from corrections beats guessing afresh in every chat.
For a hard must or must-not, say "never file an invoice over 500 euros without asking," Memory is the wrong place. Memory is a learned pattern, not an enforced rule, and it can drift over time. Firm rules belong in a Claude project or a CLAUDE.md, where you set them down deliberately and permanently instead of leaving them to a learning process.
Two layers: Memory and your own files
According to Anthropic, every Claude Project has its own Memory space, keeping knowledge from that project's chats apart from the rest. In my setup a second layer sits on top of that, and the two should not be confused.
My AI employees work in folders of their own on disk. Alongside the role description, those folders hold learning and memory files that I write into deliberately, for anything that should hold permanently. Memory in claude.ai, by contrast, picks things up from chats on its own. One layer is written and checkable, the other learned and changeable.
A typical run: I give Sonja bullet points for an email that needs to go out. She drafts a reply, and I correct any phrasing that doesn't sound like me. So that a correction does not just happen to stick, I record it in her learning file. I still read and decide before anything goes out, but the path there gets shorter with every correction I write down.
Incognito for confidential cases
For chats that shouldn't feed into Memory at all, Anthropic offers an incognito mode, reached through a ghost icon in the chat. An incognito chat isn't taken into account by Memory. That's the simplest fix when one specific conversation needs to stay without consequences, without turning off the whole Memory feature.
Memory on the Team plan: off by default
On Team and Enterprise, unlike Free, Pro, and Max, Memory is off by default according to Anthropic. The workspace owner can turn it on deliberately. For a small team, that means anyone who wants Memory has to set it up actively, it doesn't switch on automatically when you move from an individual plan to Team.
Frequently asked questions
Is Memory included in the Free plan?
Yes, according to Anthropic's help article, Memory is available on Free, Pro, and Max. Chat search, which lets you find old conversations, is reserved for paid plans.
Do I have to activate Claude Memory first?
Not on Free, Pro, or Max, where it is on by default according to Anthropic. On Team and Enterprise it is off by default and only an owner can turn it on.
How do I turn Memory off?
In settings, in the Memory section. Anthropic separates two actions there: pausing stops the learning and keeps everything already remembered, resetting deletes all memories irreversibly.
Does Memory let Claude see all my old chats automatically?
Memory brings summarized knowledge from past chats into new conversations, that's different from actively searching through old conversations. The latter is handled by the separate chat search feature, available only on paid plans according to Anthropic.
Can I export my Memory?
Yes, Anthropic offers an import and export path, but describes it as experimental itself. For anything important, rely on a Claude Project with its own files rather than a Memory export.
Does Memory store client data?
If client data shows up in your chats, Memory can store knowledge derived from it. For confidential cases, incognito mode is the safer choice, or a separate project with its own memory space per client. For how I handle client data in Claude, see Claude Privacy and GDPR: What Applies; this isn't legal advice.
Where to go from here
Look first at what Claude has already stored about you, then watch over a few weeks on a single use case such as Sonja what gets added and what doesn't. Correct wrong entries right away instead of leaving them, that's what keeps the memory useful. Why automatic recall through Memory alone isn't enough, and how fixed rules in Projects and CLAUDE.md fit in, is covered in How to Use Claude Projects Right and CLAUDE.md: Claude Code's Memory, and why classic chat tools forget by design without something like this is explained in Why ChatGPT Forgets and AI Employees Don't. The eight AI employee packages that run on exactly these mechanisms for me are available in my community: Community.
