For clinicians
Someone has handed you a page.
Here is exactly what it is, who made it, and what the tool behind it does and does not do. It takes about two minutes to read, and it is the whole story.
What the page is
A one-page agenda your patient wrote and ordered themselves. Every agenda line is verbatim. The app never paraphrases, summarises or softens what they wrote. The numbers beside each line are their own rating, one to ten, of how loud the thing was when they wrote it. Not a scale, not an instrument, and not validated as either.
What the page is not
It is not an assessment. It is not a screening result. It contains no diagnosis, no severity rating, no risk score and no clinical impression, because the app is not permitted to produce any of those. If you ever see something on one of these pages that reads like a clinical judgement, that is a defect and we would want to know.
Who controls it
The patient, entirely, and nothing reaches you that they did not put there deliberately. You do not need an account and are never asked to make one.
You may be given either of two things. A single page, which is one document and nothing else. Or a standing link, which lists every page that patient has chosen to send you and is still live — so that you are not hunting through old emails for the right one before a session.
A standing link is ongoing access, and we would rather say that plainly than imply otherwise. What it is not is access to the app: it can only ever show pages the patient added to it one at a time. Each page still expires on the schedule they chose and can be withdrawn on its own, the whole link can be withdrawn at once, and both stop working immediately. When they send you a newer page the link changes, and the previous one stops working, so the most recent thing they sent you is always the only live address.
What they can see about you reading it
Whether it has been opened. That is all. Not a count, not a timestamp beyond the first, not whether you came back to it. We deliberately do not report repeat views, because “my therapist opened it three times” is a sentence that invites an interpretation nobody can act on.
What the app does between sessions
It prompts twice a week by default, one question and never a form. The patient can set that to daily, to before-sessions-only, or off. Daily is not the default on purpose: more self-monitoring is not better for every presentation, and the most intensive option should not be the one chosen on somebody’s behalf.
The counted section, which is not the AI
A page may carry a short block of plain counts: how much of the writing is about one theme, a theme that has not appeared since a date, what the heaviest entries average against everything else, when a theme first showed up. They read like “about half of what you have written is about your mother”, and each one shows the arithmetic it rests on.
These are counted, not generated. No model is involved, the rules are fixed, and the same writing always produces the same sentences. Each one has a floor of evidence beneath it and simply does not appear below it, so a thin month produces silence rather than a confident sentence about very little.
It is off unless the patient turns it on for that specific page, and the figures are frozen at the moment they agreed to them. If they keep writing afterwards the numbers do not move, so the two of you are looking at the same document.
What the AI is allowed to do
Optionally, and off with one tap: group entries into clusters named in the patient’s own vocabulary, count what recurs, and describe a pattern in flat factual language, such as “you wrote about your mother on four days this week”. It is explicitly forbidden from naming a condition, assessing severity or risk, giving advice, interpreting cause, or addressing the patient therapeutically. Those prohibitions are enforced by three independent checks on every generation, not by asking a model to behave, and output that fails any of them is discarded whole rather than edited.
Monitoring, and why there is none
Nothing your patient writes is read by a person or scanned by a system looking for risk. There is no automated detection and no triggered intervention. This is a deliberate decision rather than a gap: detection would create a duty of response we cannot staff, and a notebook that quietly watches is worse for a patient than one that plainly does not. The app says so at sign-up, in settings, and on the footer of every sheet, alongside the 988 line.
Records, and what this is not part of
This is a consumer notebook. It is not an EHR, it does not integrate with one, and nothing here is a designated record set. What your patient shares with you is a disclosure they made, in the way they chose to make it. Whether a copy of it belongs in your own record is your professional and jurisdictional call, not ours, and we do not make a claim either way.