What one writes, the other finds.

One is better at reasoning, one at writing, one lives in the terminal and touches files. You use them for different things, often on the same day, often for the same client. The problem is not having three: it is that each knows a third of what matters, and keeps that third locked inside its own product.

The context you explain again to ChatGPT on Monday, you explain again to Claude on Tuesday. And the decision taken with Claude Code on Friday, by Monday nobody remembers.

Every AI that matters speaks MCP, the Model Context Protocol: the standard way an AI uses an outside tool. Cobrain is an MCP server. To connect an AI you paste an address into the program, authorise once on a Cobrain page, and from then on that program has the tools to search, read and write in your brain.

There is no key to copy and no secret to keep: the authorisation expires and renews itself, and you revoke it whenever you want. The guides show the exact steps for Claude, ChatGPT, Claude Code, Codex and Antigravity.

Every connected AI works with your permissions: it reads the notes you read, writes where you can write, and does not see branches that were not shared with you. There is no AI that is «more trusted» than another: the rule is applied by the database, the same for all of them and the same for the site.

And every note remembers who touched it: you from the site, Claude, ChatGPT, Claude Code. When you open a note you see who wrote it and when, and you can go back to the previous version if an AI went too far.

1. On Monday, in a chat with Claude, you agree with the client Acme that deliveries go by express courier, never on a Friday. You ask Claude to remember it: it writes a line in clients/acme/decisions.md.
2. On Wednesday, from ChatGPT, you ask for the email about next week’s delivery. ChatGPT opens the same note and writes «express courier, by the 30th, avoiding Friday». Without you having said a word.
3. On Friday, in Claude Code, you are fixing the price list. It asks the brain whether there are constraints on Acme’s prices, and finds Marta’s note saying that above a 10% discount her sign-off is needed.

Three AIs, one memory, nothing explained twice.

• It is not one single chat. The conversations stay where they are; it is the memory that sits outside.
• It does not sync the internal memories of ChatGPT or Claude: it leaves them alone. It gives the AIs a better place to look.
• It does not write on its own behind your back. It writes when your AI decides to, in a conversation with you, and every write is logged and reversible.