How to share memory between ChatGPT, Claude and Grok

Short answer

No assistant syncs its memory with another. You can copy your ChatGPT memory into Claude once, with Claude's memory import: it takes about five minutes and works on Claude's Free, Pro, Max and Team plans. After that, the two drift apart again. To keep one memory that ChatGPT, Claude, Grok and Muse all read, write it once in Upsy and connect each assistant to it. That takes about fifteen minutes.

How each assistant remembers

Two answers people look for:

• ChatGPT memory across chats: yes. With Memory on, ChatGPT uses saved memories and relevant past chats in new conversations. Temporary chats don't create memories, and an Unpersonalized temporary chat doesn't use them.
• Claude memory across devices: yes. Memory belongs to your account, so the web, desktop and mobile apps share it. Each project keeps a separate memory space, so a project chat won't know what you said outside it.

Each assistant decides what to remember. Claude's memory focuses on work and, by default, leaves out sensitive topics like health or politics; it never saves government ID numbers or financial account numbers. ChatGPT's memory summary doesn't show everything it uses. Neither shares anything with the other.

Move your ChatGPT memory to Claude

Claude has a built-in import (Anthropic's guide). It's a one-time copy, on Claude's Free, Pro, Max and Team plans, on the web and Claude Desktop.

1. Tidy ChatGPT first. In ChatGPT, open Settings › Personalization › Memory. Delete anything outdated or that you don't want to move.
2. Start the import in Claude. Open Settings › Memory and click Start import. Claude shows a prompt to copy and a box to paste the result into. (On older accounts, the Memory section is under Settings › Capabilities.)
3. Run the prompt in ChatGPT. Paste it into a normal ChatGPT chat with Memory on. A temporary chat set to Unpersonalized can't see your memories.
4. Check the answer. ChatGPT lists what it remembers about you in a code block. Read it, remove anything sensitive, and ask it whether anything is missing.
5. Paste it into Claude. Paste the list into Claude's box and click Add to memory. Claude splits it into separate memory entries you can review.
6. See what stuck. Click See what Claude learned about you. If something you need is missing, add it under Settings › Memory or tell Claude in a chat.

Anthropic calls memory import experimental. Claude may not keep personal details unrelated to work, and the copy doesn't stay in sync: what you tell ChatGPT next week won't reach Claude.

Going the other way

To move from Claude to ChatGPT, ask Claude to write out its memories of you word for word, as Anthropic suggests. Paste the result into ChatGPT and ask it to remember each point. ChatGPT saves details you explicitly ask it to remember as saved memories.

Keep one memory for every assistant

Upsy's Memory is one document you write about yourself: who you are, where you live, how you like to travel, what you eat, how you like emails written. Every assistant you connect to Upsy reads the same document.

• You write it. It's your text, up to 20,000 characters, saved as you type. Nothing is guessed or summarized for you.
• Assistants can only suggest. When you tell Claude or Grok something lasting, it can suggest adding it. The suggestion waits in Upsy until you click Add or Dismiss.
• You see who read it. The Memory page shows which assistants read it, and when.
• You choose who gets it. For each assistant, you decide whether it may use Memory.

It doesn't replace each assistant's own memory, and it doesn't write into them. An assistant reads Upsy's Memory when a task calls for it, such as booking a trip or filling in a form, or when you ask it to.

Set it up

1. Sign up free at upsy.ai with "Set up Upsy".
2. Open Memory in Upsy. A good start: paste the list you got from ChatGPT in the steps above, then edit it into what you want every assistant to know.
3. Open Assistants, click Add assistant, and pick the first one under Which assistant? Upsy shows the steps for it:
   • ChatGPT: add Upsy under Plugins as a custom MCP server (a connector that lets ChatGPT use another app) at https://upsy.ai/connect. See adding a custom MCP server to ChatGPT.
   • Claude: open Customize › Connectors, click Add custom connector, paste the same address and click Add, then Connect. Free plans get one custom connector.
   • Grok and Muse: paste the setup prompt Upsy shows into the chat, approve, and sign in. See the Grok and Muse hubs.
4. On the Upsy page that opens, sign in and turn on Memory for that assistant.
5. Paste the "remember" message Upsy shows into the assistant, so it uses Upsy in new chats without being asked.
6. Repeat step 3 for each assistant.

Check that it works

Ask each assistant the same question: "What does my Upsy memory say about how I like to travel?"

Each should read Upsy's Memory and give the same answer. Then tell one of them a new preference, like "I'm vegetarian now, remember that." It should suggest a memory, and you'll see it on Upsy's Memory page waiting for Add or Dismiss.

Prompts to try

• "Read my Upsy memory and write a short bio for a conference page."
• "Plan a weekend in Rome that fits my preferences in Upsy."
• "Draft this email the way my Upsy memory says I like emails written."
• "Suggest anything from this chat that belongs in my Upsy memory."
• "Compare what you remember about me with my Upsy memory and tell me what's different."

Troubleshooting

Claude didn't keep some of what you imported

Claude's memory favours work context and may drop personal details. Add them under Settings › Memory, or keep them in Upsy's Memory, which holds exactly what you write.

ChatGPT's export looks short

The memory summary isn't everything ChatGPT uses. Ask it for its saved memories and what it knows from past chats, and check Settings › Personalization › Memory.

An assistant ignores your Upsy memory

It may not know to look. Ask it to "check my Upsy memory", or paste the "remember" message again. Also check that Memory is turned on for that assistant on Upsy's Assistants page.

An assistant can't suggest more memories

Each assistant can have up to 20 suggestions waiting. Add or dismiss the ones on Upsy's Memory page, and it can suggest more.

Security and privacy

• Upsy's Memory is encrypted with a key of your own, according to its privacy policy. Upsy is in beta, and the policy asks you not to store confidential or sensitive data yet.
• Assistants never change your Memory directly. Every suggestion waits for you.
• Upsy's History lists each time an assistant reads your memory, under that assistant's name.
• Turning off Memory for one assistant, or turning the assistant off in Upsy, stops it reading. What it already put in its own memory stays there; delete that in the assistant.

FAQ

Can I transfer my memory from ChatGPT to Claude?

Yes. In Claude, open Settings › Memory and click Start import, run the prompt it shows in ChatGPT, and paste the result back. It's a one-time copy.

Does ChatGPT remember across chats?

Yes, when Memory is on. It uses saved memories and, if you allow it, relevant past chats. Temporary chats don't add to it.

Does Claude's memory sync across devices?

Yes. Memory is tied to your Claude account, so the web, desktop and mobile apps use the same memory.

Can ChatGPT and Claude share memory automatically?

Not with each other. Both can read the same Upsy Memory, which you keep up to date in one place.

Is it safe to share personal details across assistants?

Share only what you'd want every assistant to know. With Upsy you see who read your memory and approve every addition. For forms and bookings that use it, see Can ChatGPT fill out forms.