Memory you can
see.

You explain yourself to every AI, over and over, and none of them remember. TotalRecall keeps your context in one place you own. Start a project on Claude, finish it on ChatGPT — they both know what you decided. Grant access with a drag. Take it back the same way.

The memory engine is atomir — MIT-licensed, open source, and yours to read, fork, or self-host. TotalRecall adds the permission surface on top.

Your context is trapped, and nobody has to fix it but you.

Portable memory is bad for retention, so no incumbent will build it. Which leaves three problems you live with every day.

• 01

  You repeat yourself constantly.

  Every new chat starts from zero. Every new tool means re-explaining your work, your preferences, your project. Again.

• 02

  Your context is locked in.

  ChatGPT's memory stays in ChatGPT. Claude's stays in Claude. Switching tools means starting over — which is exactly how they want it.

• 03

  You can't see what they know.

  Memory happens to you, silently. You didn't decide what was saved, you can't audit it, and you can't take it back.

Three steps. Nothing to install.

TotalRecall runs as a memory layer your assistants connect to through their own connector settings. Everything sits on your board, and you decide who sees what.

• 01

  Connect.

  Link your assistants in a couple of clicks — Claude and ChatGPT, plus the coding tools you already live in (Claude Code, Cursor, Codex, Gemini CLI). No browser extension, no session hijacking, no scraper.

• 02

  Decide.

  Drag a provider onto a project to let it read and write. Drag it off to cut it out. Takes effect on the next request, no re-auth, no delay.

• 03

  Continue.

  Make a decision on one assistant, pick it up on the other. Your memory follows you, because it was never theirs.

Start where you left off.

Your history doesn't have to stay in the tools that trapped it. Upload your official ChatGPT or Claude export archive and TotalRecall reads it — pulls out the facts you shared, the decisions you made, the preferences you kept restating — and drops them into your inbox as candidate cards, grouped by where they came from.

Nothing goes live until you accept it. Nothing is shared with any assistant until you grant a space. Your board doesn't start empty; it starts familiar.

Every read. Every write. Receipted.

Ownership without a paper trail is a claim. This is the paper trail: which assistant read which cards, which assistant wrote what, when. You see all of it — and you can't quietly overwrite it, because you can't delete the log yourself. That's the point.

Move it. Rewrite it. Mark it private.

Every fact, decision, or preference is a card. Cards sit in spaces — one per project. Move a card. Rewrite it. Delete it. Mark it private and no assistant will ever see it — not even the one that wrote it.

Each card carries its origin — typed by you, from your ChatGPT export, written by Claude. You never wonder how something got in.

This product is about your data. So we treat it that way.

• Your memory is stored on Google Cloud, encrypted at rest. Any AI key you add is encrypted separately and never shown again.
• Nothing is shared with any assistant unless you grant it, per project.
• Delete any memory from your board, any time. Ask, and we delete your whole account.
• We don't stream your conversations. We only see what your assistant deliberately saves, or an export you choose to import.
• Your memory is not used to train any model.

One memory. Every AI. You hold the keys.

TotalRecall is in open beta. Create an account, connect Claude or ChatGPT, and watch the first memory land on your board. Free, no card.