Claude memory
Claude memory: what it remembers, and what it can’t.
Claude can remember you across chats now, and Projects keep files and instructions together. For many people that’s enough. Here’s what each kind of memory is good at, where it stops, and when a shared vault that every assistant can read is worth the five minutes.
Reading time: 9 minutesUpdated:
01
Three kinds of memory, three different jobs
They’re often lumped together as “AI memory”, but they solve different problems.
Built-in memory
Claude picks up what matters from your chats: your role, your projects, how you like answers. It writes this down itself, and brings it into new chats.
Projects
A workspace with its own files, its own instructions and its own memory. Good for one topic you keep coming back to: a client, a thesis, a launch.
Shared vault over MCP
Notes in a place outside Claude that every connected assistant reads and writes. You decide the structure, and every change is visible.
02
Claude’s built-in memory: what it does well
Claude can carry context from one chat to the next. It builds a summary of what it has learned about you, such as your job, your current projects and your preferences, and uses it in new conversations. You can also ask it to look up what you discussed in earlier chats. In the settings, you see what Claude remembers, you can edit it or switch memory off, and incognito chats aren’t remembered at all. What exactly is available depends on your plan.
Projects go one step further. Each project has its own instructions, its own files and its own memory, kept apart from your other chats. So your client work doesn’t leak into your private questions, and a long-running project has a home.
Claude Code works differently again: it reads CLAUDE.md files in your project and home folder, a kind of memory as plain text that you write yourself.
The big strength of all three: zero setup. You don’t have to think about structure; Claude does the remembering. For “know that I’m a designer and prefer short answers”, that’s exactly right.
03
Where built-in memory stops
It only lives in Claude. ChatGPT, Cursor or Gemini don’t see it. Each of them builds its own memory of you, and the copies drift apart. You can carry memories over from one assistant to another, but they won’t stay in sync.
It remembers you, not your knowledge. A memory summary is great for preferences and context. It’s not built for a decision with its reasons and the options you rejected, a project’s history over six months, or twenty sources on a topic you’re researching.
Claude decides what to keep. You can correct the summary, but its shape and wording come from the model, not from you. And you see the current state, not a history of what changed when.
It isn’t a set of files. Your memory stays in your Claude account. Useful there, but not something you can open in another app, link to or hand to a colleague.
04
Claude’s memory or a shared vault?
The two don’t compete. They hold different things.
| – | Claude’s memory & Projects | Shared vault (WhyVault over MCP) |
|---|---|---|
| Setup | None, it’s built in | Connect once with a URL, about five minutes |
| Who writes it | Claude picks what to keep; in Projects you add files and instructions | You and your agents, in notes you structure |
| Which tools see it | Only Claude; a project’s memory only inside that project | Every MCP client: Claude, ChatGPT, Cursor, Claude Code and more |
| Best at holding | Preferences, context about you, the thread of recent chats | Decisions with reasons, project histories, sources, knowledge you build on |
| Seeing changes | The current state | Every change as a version with the agent’s name, undo in one click |
| Format | Inside your Claude account | Plain Markdown, export as a zip anytime |
| Where it lives | With Anthropic, under your Claude plan’s terms | EU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too |
| Cost | Included in Claude | Free up to 20 notes, Pro €9.99 a month |
- Setup
- Claude’s memory & ProjectsNone, it’s built in
- Shared vault (WhyVault over MCP)Connect once with a URL, about five minutes
- Who writes it
- Claude’s memory & ProjectsClaude picks what to keep; in Projects you add files and instructions
- Shared vault (WhyVault over MCP)You and your agents, in notes you structure
- Which tools see it
- Claude’s memory & ProjectsOnly Claude; a project’s memory only inside that project
- Shared vault (WhyVault over MCP)Every MCP client: Claude, ChatGPT, Cursor, Claude Code and more
- Best at holding
- Claude’s memory & ProjectsPreferences, context about you, the thread of recent chats
- Shared vault (WhyVault over MCP)Decisions with reasons, project histories, sources, knowledge you build on
- Seeing changes
- Claude’s memory & ProjectsThe current state
- Shared vault (WhyVault over MCP)Every change as a version with the agent’s name, undo in one click
- Format
- Claude’s memory & ProjectsInside your Claude account
- Shared vault (WhyVault over MCP)Plain Markdown, export as a zip anytime
- Where it lives
- Claude’s memory & ProjectsWith Anthropic, under your Claude plan’s terms
- Shared vault (WhyVault over MCP)EU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too
- Cost
- Claude’s memory & ProjectsIncluded in Claude
- Shared vault (WhyVault over MCP)Free up to 20 notes, Pro €9.99 a month
You don’t have to choose. Leave Claude’s memory on for the small things, and keep what has to last, and what other tools need too, in the vault.
05
When is which enough?
A quick way to decide:
Built-in memory is enough if …
you only use Claude, and mainly want it to know who you are, what you’re working on and how you like answers.
Add Projects if …
you keep coming back to a few clearly bounded topics with fixed material: a client, a course, a book, a launch.
A shared vault makes sense if …
you use more than one AI tool, want decisions and reasons kept in your own words, need to see and undo what an agent changed, or want your knowledge in open files.
06
Using all three together
In practice, the best setup splits the work: Claude remembers you, the vault remembers your work.
- 1
Leave built-in memory on for the small things
Tone, preferences, what you asked about yesterday. That’s what it’s built for, and it costs you nothing.
- 2
Use a Project for a bounded job
Upload the brief, the contract or the style guide. In the project instructions, add one line: “For anything about my projects and decisions, check the WhyVault connector first.”
- 3
Connect your vault once
Add WhyVault as a custom connector in Claude, with the URL
https://mcp.whyvault.io. It then works on the web, on desktop and in the mobile app, and the same URL connects ChatGPT and Cursor. Connect Claude - 4
Write down how memory should work
A short rules note in
9_Systemtells every connected agent what belongs in the vault. Every assistant reads it at the start of a session.# Memory rules - Before you answer anything about my projects, search the vault. - When we decide something, write a decision note in the right room: what we decided, why, and what we rejected. - Preferences about how I like things written go in 9_System/Identity/SOUL.md, not only in your own memory. - Tell me in one line what you saved.
07
What it feels like
Yesterday you made a decision with Claude. Today you’re in ChatGPT:
You: What did I decide with Claude yesterday about the newsletter?
- Tool: search_notes · “newsletter decision”
ChatGPT: In “Newsletter – weekly instead of daily”, written by Claude yesterday: weekly on Thursdays, because the daily issue took two hours a day and open rates were falling. Rejected: a monthly digest, too far from the news.
You: Good. Draft the first Thursday issue on that basis.
- Tool: read_note · “Newsletter – topics”
ChatGPT: Here’s a draft with the three topics from your list. Want me to save it as a note in the room?
Memory inside one assistant belongs to that assistant. A vault belongs to you.
?
Frequently asked questions
Does Claude have memory?
Yes. Claude can remember context across chats, look up earlier conversations and keep a separate memory per project. In the settings you can see and edit what it remembers, or turn memory off. Which features you get depends on your plan.
What’s the difference between memory and Projects?
Memory works across all your chats and is written by Claude. A project is a workspace you set up for one topic, with your own files and instructions, and a memory of its own that stays inside that project.
Can ChatGPT or Cursor read Claude’s memory?
No. Each assistant keeps its own memory. A vault connected over MCP is the shared layer: Claude, ChatGPT, Cursor and Claude Code all read the same notes. What is MCP?
Do I have to turn off Claude’s memory to use WhyVault?
No. They work side by side. Claude’s memory keeps the small things about you; the vault keeps decisions, projects and knowledge that should last and be visible to other tools.
Where is my vault stored?
WhyVault is hosted in the EU, in Frankfurt, and GDPR-compliant; businesses get a DPA. What an agent reads from your vault is also processed by its provider, for Claude that’s Anthropic. History, export and privacy
How much does it cost?
Claude’s memory is part of your Claude plan. WhyVault is free for up to 20 notes, no credit card needed. Pro costs €9.99 a month or €99 a year, with no limit. Reading and exporting always work, even without a subscription.
Keep reading
All guidesMCP memory server
What a memory server for Claude, ChatGPT and Cursor does, local vs hosted, and what to look for before you pick one.
ReadLLM Wiki
Karpathy’s idea explained: an AI maintains your knowledge as a Markdown wiki. With a template to copy and a hosted setup, no scripts.
ReadDecision log
A template for decisions you’ll still understand six months from now – with a filled-in example to copy.
Read
One memory for every assistant.
Connect Claude to your vault in five minutes, then ChatGPT and Cursor with the same URL. Free, no credit card.