Short answer
In Claude, open Settings, Memory, Start import, run the prompt it gives you in ChatGPT, and paste the reply back. That moves a summary of you, not your chats, Project files or anything your team shares. For work that should survive the next switch too, keep it in markdown notes both assistants read over MCP.
To transfer your ChatGPT memory to Claude, open Claude's Settings, then Memory, and choose Start import. Claude shows you a prompt to run in ChatGPT. You paste ChatGPT's answer back into the box and Claude files it as memory. That takes about a minute and it works. What it carries over is narrower than most guides suggest, and the gap matters most if you used ChatGPT for work other people depend on.
The import moves a description of you: how you like answers formatted, what you do, which tools you use, what you are working on. It does not move the work itself. The decisions, the drafts, the reference files and the long threads where something got figured out all stay in your ChatGPT account. Baalda comes in for that second half. Baalda keeps that kind of knowledge as plain markdown files on your own disk, and both assistants can read those files over the Model Context Protocol. Then switching assistants doesn't mean moving your work at all.
How do you transfer ChatGPT memory to Claude?
Anthropic's help centre describes three steps, and as of 13 October 2026 they are:
- In Claude, go to Settings, then Memory, and select Start import. The import screen shows a prompt and a text box.
- Copy the prompt into a chat with ChatGPT. It is written to make ChatGPT write out everything it has stored about you in one reply.
- Paste that reply into Claude's text box and click Add to memory. Claude splits it into individual memory edits, which you can review under Manage edits.
Anthropic lists memory import for the Free, Pro, Max and Team plans, on the web and in Claude Desktop. The same page says the update can take up to 24 hours to show, and that the feature is "experimental and still in active development", so Claude "may not always successfully incorporate imported memories". It also says Claude's memory focuses on work, so it may not keep imported personal details unrelated to work.
Check the result before you trust it. Open Claude's memory afterwards and read what it actually kept. Anything wrong or missing is easier to fix now than after a month of answers built on it.
What does the import actually move, and what stays in ChatGPT?
The whole import is a block of text that ChatGPT writes about you on request. Nothing in the flow opens your chat history, your files or your settings. If ChatGPT never stored something, or leaves it out of that one reply, Claude never sees it.
| Stays in ChatGPT unless you move it | Why the import misses it |
|---|---|
| Your conversation history | The import reads one reply, not your threads |
| Files in Projects and uploads | The prompt asks for memories, not attachments |
| Custom GPTs and their instructions | They are configuration, not memory |
| Anything a teammate also relies on | The memory describes one person |
The last row matters most. If you were the person on your team who kept the onboarding answers, the pricing logic or the client history in a ChatGPT Project, the import moves a sentence like "works on client onboarding" and leaves the onboarding behind.
Can you export your ChatGPT data first?
Yes, and do it before you cancel anything. In ChatGPT, open Settings, then Data controls, and choose Export under Export data. OpenAI's help centre says the export arrives by email or text and can take up to 7 days, and the download link expires 24 hours after you receive it. The archive holds your chat history as conversations.json plus a readable chat.html.
Two limits are worth knowing before you count on it. Self-service export covers ChatGPT Free, Go, Plus, Pro and eligible Edu workspaces. In a ChatGPT Business, Enterprise or Healthcare workspace, OpenAI says to ask the workspace owner, because self-service export is not available there. And the archive is your history, not a working document. A JSON file of every thread you ever had does not tell the next assistant what is currently true.
Saved memories have no export button of their own. You can read them under Settings, Personalization, Memory, which is also the place to check that the reply you pasted into Claude was complete.
Why move the work out of both assistants instead?
Because you just did this once, and the import put everything back into one vendor's memory store. If you switch again next year, or your team standardises on something else, you will run the same prompt in reverse and lose the same things again.
The alternative is to keep what has to survive in files that neither assistant owns. Baalda is built for that. A vault is a folder of .md files on your own disk. It syncs to your team in real time, and the server exposes it over MCP at one endpoint, POST /api/mcp. Claude connects to it as a custom connector. ChatGPT connects to the same server as a custom MCP connector over OAuth. Both call the same tools, search_notes to find something and read_note to open it, and both see the same files.
So the migration you just did becomes a different kind of task. Take the reply ChatGPT wrote for Claude's import and save it as a note as well, for example Context/how-i-work.md. Then go through your Projects and pull out the parts somebody else would need. Make them notes too: Clients/acme.md, Decisions/pricing-2026.md. You can do this by hand, or ask the assistant to write them with create_note, which writes a real markdown file at the path you name.
For Claude Code, which can already read the disk, the connection is one command, using a token from Vault Settings, MCP:
claude mcp add --transport http context https://api.baalda.com/api/mcp \
--header "Authorization: Bearer mcp_…"Once that is done, the next switch costs one connection, because the files never belonged to either assistant. The setup and the OAuth flow are in the MCP setup guide. The same argument for models rather than products is in notes outlive the model.
What changes when it is a team switching, not one person?
Claude's memory import is per person, and so is ChatGPT's memory. A team that moves from ChatGPT to Claude does the import once per seat and still ends up with no shared copy of anything. Each person's assistant knows a little about that person and nothing about the team.
A shared vault reverses that. Every member's assistant connects to the same vault with that member's own token, so each one sees exactly what that person is allowed to see, and a change one person's assistant makes is a change to a file everyone can open. Someone can correct a note that is wrong. Version history shows what changed and lets you restore an earlier version. Nobody has to export their memory when they leave, because the useful part was never in their memory.
If your team is mid-switch, with some people still on ChatGPT and some on Claude, this is also how they share context in the meantime. Both groups read the same Decisions/ folder from different assistants.
Where does this not help?
Be clear about the limits before you set it up.
- Baalda does not import your ChatGPT export. There is no converter for
conversations.json. Going through old threads and keeping the parts worth keeping is still a person's job, or a job you ask an assistant to do one thread at a time. - Preferences belong in Claude's memory, not in a note. "Answer in British English" and "skip the preamble" are what Claude's memory import is for. Use it for that.
- ChatGPT cannot reach a vault on your laptop. Baalda runs free and local, but chatgpt.com cannot see
localhost. ChatGPT needs the managed service or your own deployment at a public HTTPS address, which is covered in using ChatGPT as a second brain. - Claude's connector limits apply. Anthropic says Free users get one custom connector, and on a Team plan only Owners and Primary Owners can add one.
- For one person who only used ChatGPT for themselves, the import may be the whole answer. A vault earns its place when somebody else needs to read the same knowledge, or when you expect to switch again.
If you are choosing what Claude should read in the first place, the best second brain for Claude compares the options, including the ones that are not Baalda.