Canonly keeps track of what you decided, what changed and what you ruled out, so ChatGPT, Claude and your other AI tools stop working from information you have already replaced.
Your project already lives in a repository, a drive, a folder your AI writes into and a dozen chats. Keep it there. Canonly works out what is current, what was replaced, what you ruled out and which documents disagree, and gives every AI you use the current answer.
ChatGPT, Claude and Cursor write notes, plans and decisions. Later, a different AI reads an old one and answers from it confidently, with no sign that it was replaced. Canonly marks one version current and the rest replaced, notices when two of your documents disagree, and remembers what you turned down and why.
Connect a GitHub repository, drag in a folder, or import from Google Drive. Any AI that speaks MCP can then ask what is current instead of guessing.
Not only what you agreed to. What you ruled out, and why, which is the part nothing else keeps, because the thing you refused never became a document.
Most tools hand back everything they have ever seen and leave you to work it out: Claude said £12, ChatGPT said £15, your notes say £10, last year’s plan says £9. Four answers, and you decide which one is right.
Canonly returns one. £12 is current and you confirmed it. £15 you turned down. £10 and £9 were replaced, and are kept as history rather than served as the answer. Every AI you use gets the same one.
When an AI proposes something you have already refused, Canonly answers with the decision you made, the date you made it and the reason you gave. A memory feature cannot do this: it either stores a fact and serves it as though it were true, or discards it and meets the same wrong idea as new next month. A refusal is neither, so it is usually thrown away.
A project folder holds files and gives them to one assistant. It does not record which of those files is still true, does not tell you when two of them disagree, and does not survive the move to a different AI tool. Canonly sits above whichever tools you use and answers all of them. The longer comparison goes through it question by question.
Skills are pieces of AI work that read your project: meeting notes into decisions, a paper against your findings, what a change makes out of date. They propose; you decide what counts. A skill never writes to a document on its own.
Every project gets a check that names what is missing before an AI trips over it, and offers something to do about each gap. It is free, and it reaches a verdict without calling a model.
Don’t start with a blank page. Templates for software, research, dissertations, startups and consulting set up the documents your AI needs, and you write in plain rich text, with no Markdown required.
The whole core loop is free forever: connect a source, mark what is current, approve it, see what conflicts, and let your AI read the trusted answer. Public knowledge bases are free and unlimited. Create an account, or read what the paid plans add.
