Your AI reads your Desk. That is the short version, and this page is that sentence made specific: what it reads, why, what it never does with it, and exactly what a person can and cannot see.
I am writing this because the people using it asked for it. They kept running into an assistant that told them a card was waiting when they had already handled it. That is not a small annoyance. It is the difference between something you can rely on and something you have to double-check, and once you are double-checking it, it is costing you time instead of giving it back.
One more thing you should know about this page: it explains, it does not govern. Your signed agreement comes first, then the Terms of Use, then this page. If this page ever differs from either of those, they win, in that order.
Why it has to see
Your Desk is a page. What your AI knows about it used to come from its own records of what it had done, rather than from the page you are actually looking at. Most of the time those agree. When they drift apart, you get told something confidently untrue about your own work.
An assistant that cannot see your Desk will eventually tell you a card is waiting when you already handled it, or that something is done when it is not. So it reads the Desk and checks itself against what is really there before it tells you anything about it. You can only see if you have eyes.
What it reads, and what it is for
It reads your Desk: the cards on it, whether they are open or done, the messages on them, and what the page actually shows. It reads that to check itself — to confirm that what it is about to tell you matches what is there, that something it placed actually arrived, that an action you took actually landed.
It is not for building a profile of you, training a model on your material, marketing, or resale. Your material is yours. It is never reused for another client and never for my own promotion.
What a person can and cannot see
I want to be straight here, because a promise nobody can keep is worse than no promise.
No person is reviewing your Desk as a matter of routine. The reading is the AI checking its own accuracy.
But the AI works with me, and I can ask it about your work. So the honest line is not "no human will ever see anything." It is this: I do not browse your Desk. I see your material when you raise something, when you ask for help, when something breaks and needs a person, or when I am doing the work you hired me for — the same access any consultant has to the work you handed them. And, rarely, when something you post needs a person's judgment, for your safety or because the law draws a line an AI must not cross alone, the AI first tells you what it can and cannot do, and a copy reaches me so a person is accountable for what happens next.
If you want a harder boundary than that, say so and we write it into your agreement, and that line wins, except where the law or your safety requires a person's judgment. Please do not assume it.
What I never touch
I never ask for your passwords. Access to your own accounts is something you grant from your side, and you can revoke it yourself, in one step, without changing a single password.
Nothing goes out in your name without your say-so, at whatever level you have set. Everyone starts at draft-only: the AI writes, you send.
Nothing behind a door you did not open. Your Desk, and what you choose to connect. That is the whole surface.
Honest limits
The AI is good and it is not infallible. Where it cannot verify something, it is built to say so rather than guess. If it ever tells you something about your Desk that turns out to be wrong, that is a defect — I want to hear about it, and the fix goes into how it works, not into that one card.
If anything here is unclear, ask me. I would rather explain it than have you agree to something you did not read.