Moving your whole AI history to a new model — and actually owning it.

People are hand-migrating years of ChatGPT into Claude with 100MB export files. There is a better way to do it — and a reason you should only ever have to do it once.

Search “move my ChatGPT history to Claude” and you will not find a debate. You will find instructions. Reddit threads with step counts. YouTube walkthroughs. Substack guides with screenshots. Settings, Data Controls, Export, wait for the email, download the zip, find the chat.html file that is somehow 100 megabytes, then try to feed it to a model with an upload limit smaller than the file.

That is not a niche complaint. That is a migration happening by hand, right now, by people who have decided their history is worth the afternoon.

Why it's this hard

Both platforms have memory. ChatGPT remembers useful context from your chats. Claude remembers you too. The memory is real, and it works — inside the walls it was built in. That is the whole catch. ChatGPT's memory of you is not something you can hand to Claude, and Claude's is not something you can hand to anything else.

So the export exists, and the guides exist, because the portability doesn't. The vendors will remember you. They just won't let the memory leave with you.

What you're actually trying to move

Here is the thing worth being precise about: you are not trying to move a transcript. Nobody wants to reread three years of chat logs. What you want to carry across is the through-line — the problems you keep returning to, the decisions you already made and why, the questions you circle, the thread you set down six months ago and never picked back up.

That is the part with compounding value. And it is the part a 100MB HTML file technically contains and practically buries.

The version that takes a minute

Export your data the normal way — ChatGPT or Claude, both work. Then drop the zip on export-rescue. It reads the archive in your browser, on your machine. Nothing is uploaded, because there is no server in the path to upload it to.

What comes back out is not a transcript. It is a vault: one note per conversation, wikilinked, with topic hubs assembled across the whole archive so the threads are actually visible. Plain markdown files. Open them in Obsidian, or a text editor, or grep them. They sit on your disk. They are yours in the boring, literal sense — you can delete us and keep them.

Why you only do this once

The reason to do the migration properly is that it is the last one. Once your history is files you own, the model stops being the thing that holds your memory and goes back to being what it should have been all along: a rented engine you point at your own material.

A better model ships next quarter — and it will. You do not migrate. You point it at the vault. The switching cost that the platforms are quietly counting on collapses to nothing, because the thing that made switching expensive was never the model. It was your memory being held hostage inside it.

Your AI history is not a backup. It is the record of how you think. It should not live somewhere you cannot follow it.