ChatGPT memory
ChatGPT memory: share it, don’t re-explain it.
ChatGPT remembers a lot about you, but only for itself. The moment you move on to Claude or Cursor, you start from scratch. Here’s how ChatGPT’s memory works, where it stops, and how to set up one memory that all three can read.
Reading time: 9 minutesUpdated:
01
How ChatGPT remembers things
“ChatGPT memory” covers several features. Each one does a different job.
Saved memories
Short facts about you that ChatGPT keeps, for example when you say “remember that …”. You see them as a list in the settings and delete single ones or all of them.
Referencing past chats
ChatGPT can draw on earlier conversations without you saving anything. Handy for picking up the thread – but it doesn’t remember every detail.
Projects
A workspace with its own chats, files and instructions. Depending on the setting, a project only draws on the chats inside it.
Custom instructions
A fixed text ChatGPT follows in every chat: who you are, how you like answers. You write it yourself, and it doesn’t change on its own.
02
What ChatGPT’s memory does well
ChatGPT has a real, built-in memory. Saved memories hold what you explicitly want kept: your job, your family, that you hate bullet points. On top of that, ChatGPT can reference past chats and pick up a conversation from last week. You switch both on or off separately in the settings, and temporary chats aren’t remembered at all. How much of this you get depends on your plan.
Projects bundle chats, files and instructions around one topic. If you limit a project to its own memory, client A doesn’t leak into your questions about client B. Custom instructions are the simplest form: a text you write once that applies to every chat.
The big strength: zero setup. You don’t have to come up with a structure. For “remember that I’m a teacher and need examples from primary school”, that’s exactly right, and it costs you nothing extra.
03
Where ChatGPT’s memory stops
It stays in ChatGPT. Claude, Cursor or Claude Code don’t see any of it. Each assistant builds its own picture of you, and the copies drift apart. You can copy memories out of ChatGPT and paste them elsewhere, but after that they won’t stay in sync.
You only see part of it. Saved memories show up as a list in the settings. What ChatGPT pulls in from past chats you don’t see as a list, and there’s no history of what changed when.
It remembers you, not your work. A memory is a sentence. It’s not built for a decision with its reasons and the options you rejected, a project’s history over months, or the sources on a topic.
It isn’t a set of files. You can export your data from OpenAI, but you don’t get notes you can open in another app, link to, or that Claude reads directly.
04
ChatGPT memory or a shared vault?
The two don’t rule each other out. They hold different things.
| – | ChatGPT memory & Projects | Shared vault (WhyVault over MCP) |
|---|---|---|
| Setup | None, it’s built in | Connect once in developer mode, about five minutes |
| Who writes it | ChatGPT picks what to remember; in Projects you add files and instructions | You and your agents, in notes you structure |
| Which tools see it | Only ChatGPT; a project with its own memory only that project | Every MCP client: ChatGPT, Claude, Cursor, Claude Code and more |
| Best at holding | Preferences, facts about you, the thread of recent chats | Decisions with reasons, project histories, sources, knowledge you build on |
| Seeing changes | The list of saved memories, no history | Every change as a version with the agent’s name, undo in one click |
| Format | Inside your ChatGPT account | Plain Markdown, export as a zip anytime |
| Where it lives | With OpenAI, under your ChatGPT plan’s terms | EU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too |
| Requirements | Depends on your plan, a lighter version on the free plan too | In ChatGPT from Plus (developer mode, in the browser); nothing extra in Claude and Cursor |
| Cost | Included in ChatGPT | Free up to 20 notes, Pro €9.99 a month |
- Setup
- ChatGPT memory & ProjectsNone, it’s built in
- Shared vault (WhyVault over MCP)Connect once in developer mode, about five minutes
- Who writes it
- ChatGPT memory & ProjectsChatGPT picks what to remember; in Projects you add files and instructions
- Shared vault (WhyVault over MCP)You and your agents, in notes you structure
- Which tools see it
- ChatGPT memory & ProjectsOnly ChatGPT; a project with its own memory only that project
- Shared vault (WhyVault over MCP)Every MCP client: ChatGPT, Claude, Cursor, Claude Code and more
- Best at holding
- ChatGPT memory & ProjectsPreferences, facts about you, the thread of recent chats
- Shared vault (WhyVault over MCP)Decisions with reasons, project histories, sources, knowledge you build on
- Seeing changes
- ChatGPT memory & ProjectsThe list of saved memories, no history
- Shared vault (WhyVault over MCP)Every change as a version with the agent’s name, undo in one click
- Format
- ChatGPT memory & ProjectsInside your ChatGPT account
- Shared vault (WhyVault over MCP)Plain Markdown, export as a zip anytime
- Where it lives
- ChatGPT memory & ProjectsWith OpenAI, under your ChatGPT plan’s terms
- Shared vault (WhyVault over MCP)EU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too
- Requirements
- ChatGPT memory & ProjectsDepends on your plan, a lighter version on the free plan too
- Shared vault (WhyVault over MCP)In ChatGPT from Plus (developer mode, in the browser); nothing extra in Claude and Cursor
- Cost
- ChatGPT memory & ProjectsIncluded in ChatGPT
- Shared vault (WhyVault over MCP)Free up to 20 notes, Pro €9.99 a month
You don’t have to choose. Leave ChatGPT’s memory on for the small things, and keep what has to last, and what Claude and Cursor need too, in the vault.
05
When is which enough?
A quick way to decide:
ChatGPT’s memory is enough if …
you almost only use ChatGPT, and mainly want it to know who you are 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 job application.
A shared vault makes sense if …
you use ChatGPT alongside Claude or Cursor, want decisions kept in your own words, need to see what an agent changed, or want your knowledge in open files.
06
Set up one memory for ChatGPT, Claude and Cursor
You connect each tool once, with the same URL. After that, they all read and write the same vault.
- 1
Create a WhyVault account
Sign up for free. Your vault starts with a ready-made structure and the
9_Systemfolder, where the rules for your agents live. See the structure - 2
Turn on developer mode in ChatGPT
On chatgpt.com in the browser, turn on developer mode in the settings. It’s available from Plus (also Pro, Business, Enterprise and Edu), not on the free plan and not in the ChatGPT apps. In company workspaces, an admin decides.
- 3
Add WhyVault in ChatGPT
Create a new app with the URL
https://mcp.whyvault.ioand OAuth as authentication, then sign in to WhyVault. In a new chat, pick WhyVault from the tools menu. ChatGPT asks before every write. Connect ChatGPThttps://mcp.whyvault.io
- 4
Connect Claude and Cursor with the same URL
In Claude, add WhyVault as a custom connector; in Cursor, put the URL in
mcp.json. Both sign in via OAuth, no key needed. Connect Claude · Connect Cursor{ "mcpServers": { "whyvault": { "url": "https://mcp.whyvault.io" } } } - 5
Set permissions per connection
Under Connect agents you see every connection and decide whether it only reads or also writes. Deleting stays off until you allow it. History, export and privacy
# 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.- When I say “hand this over”, sum up the chat as a note: goal, status, open questions, next step.- 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
The rules note: write it once, every agent reads it
Instead of maintaining the same instructions in ChatGPT, Claude and Cursor, you write them once in the vault. Every connected agent reads them at the start of a session through vault_overview.
One place for all. Change a rule here, and it applies in ChatGPT, Claude and Cursor from the next session on.
Handover as a command. “Hand this over” turns a long chat into a short note the next tool can find.
Keep custom instructions short. In ChatGPT, one line is then enough: “At the start, read my WhyVault and follow the rules there.”
08
What the handover feels like
In the morning you worked through a pricing structure in ChatGPT. In the afternoon you build the page in Claude:
You: What did I settle on for pricing in ChatGPT this morning?
- Tool: search_notes · “pricing handover”
Claude: In “Pricing – handover”, written by ChatGPT this morning: two tiers instead of three, a yearly price with two months off. Still open: whether there’s a trial period.
You: Good. Turn that into the pricing table for the landing page.
- Tool: read_note · “Pricing – decisions”
Claude: Here’s the table with both tiers. I added the open question about the trial at the bottom of the note so you don’t lose it.
09
Three workflows that hold up
Each tool does what it’s best at. The vault keeps the thread.
Pass context along
At the end of a chat, say “hand this over”. The agent writes the goal, status and next step into a note. In the next tool, ask: “Where were we with …?”
Keep decisions
When you decide something, the agent creates a decision note: what, why, what was rejected. Months later, every assistant finds the reasons. Decision log
Think in ChatGPT, build in Cursor
In ChatGPT you work out the requirements, which land as a note in the project room. Cursor reads them while you code and adds what changed in the code.
Memory inside one assistant belongs to that assistant. A vault belongs to you.
?
Frequently asked questions
Does ChatGPT have memory?
Yes. ChatGPT can keep saved memories about you, draw on past chats and use separate files and instructions per project. In the settings you switch these on or off, see your saved memories and delete them. What exactly you get depends on your plan.
Can I share ChatGPT’s memory with Claude?
Not directly. Each assistant keeps its own memory. You can copy memories and paste them into Claude, but they drift apart after that. A vault connected over MCP is the shared layer: ChatGPT and Claude read and write the same notes. More on Claude memory
Do I need a paid ChatGPT plan for this?
Yes, for the connection to WhyVault. ChatGPT only adds your own MCP servers in developer mode, which starts with Plus, in the browser on chatgpt.com. In Business, Enterprise and Edu workspaces, an admin has to allow it. Connect ChatGPT
Do I have to turn off ChatGPT’s memory?
No. They work side by side. ChatGPT’s memory keeps the small things about you; the vault keeps decisions, projects and knowledge that should last and be visible to Claude and Cursor.
What does OpenAI see of my vault?
Whatever ChatGPT reads from your vault is also processed by OpenAI. WhyVault itself is hosted in the EU, in Frankfurt, and GDPR-compliant; businesses get a DPA. If in doubt, set the connection to read only. History, export and privacy
Can I undo what ChatGPT writes to the vault?
Yes. Every change becomes a version with the connection’s name. You restore an earlier version in one click, or roll back everything a connection changed since a point in time.
How much does it cost?
ChatGPT’s memory is part of your ChatGPT 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 guidesClaude memory
Built-in memory, Projects or a shared vault over MCP: an honest comparison, and when each one is enough.
ReadMCP memory server
What a memory server for Claude, ChatGPT and Cursor does, local vs hosted, and what to look for before you pick one.
ReadClaude connectors
What connectors in Claude are, how to add a custom one with an MCP URL, and which ones are actually worth it.
Read
One memory for every assistant.
Connect ChatGPT to your vault in five minutes, then Claude and Cursor with the same URL. Free, no credit card.