Our limits
Our limits
This category has a bad record. These are the commitments we put in writing — and how we verify them.
We never diagnose
Nor confirm a suspicion. If you ask whether you have an illness, the answer is that we do not speak about that — and that the question belongs to a doctor.
Practising medicine without a licence includes making a diagnosis.
How we verify it
A classifier runs on every message, in both directions. An answer that crosses the line is blocked before it reaches you.
We never promise a cure
Of anything, not even by symbolic means. There is no card, transit or ritual that cures illness.
Advertising a cure by secret or infallible means is a crime in Brazil — including in the advertising itself.
How we verify it
Every marketing piece is reviewed before it goes up. The block at generation time is hard, not a tone adjustment.
We never prescribe substances
Tea, herb, bath, tincture, formula. Nothing with therapeutic intent, not even something harmless.
Case law reaches even harmless substances: the recognised harm is steering someone away from proper treatment.
How we verify it
A block on the model output, and immediate removal for any specialist who does it.
We never discourage treatment
If you are in therapy, on medication or under care, we encourage you to continue. Always.
Interrupting treatment on our advice would be exactly the harm the law tries to prevent.
How we verify it
The pattern is detected both in your message and in the answer. When the subject appears, an explicit notice comes with it.
We never sell fear
There is no "path opening", "cleansing" or "binding" for sale here. A hard forecast is delivered with care and agency, never as a threat followed by an offer.
Using influence to obtain financial advantage is fraud.
How we verify it
It is the most documented abuse pattern in this category abroad. We audit conversations by sampling and remove specialists on the first occurrence.
A crisis stops the reading
If a signal of risk to life appears, the oracular conversation stops. No card is drawn, no transit is read. You get support resources and an invitation to come back whenever you want.
How we verify it
Detection happens before any call to the model. The response is fixed and written by people, not generated.
Why we ask for two consents
The content of your conversations may reveal philosophical or religious belief, and data protection law treats that as sensitive personal data. For sensitive data, the ordinary legal bases do not apply — only the stricter ones, which in practice require specific and highlighted consent.
That is why sign-up has two boxes, not one. Burying that consent in a general acceptance of the terms is the shortcut almost the whole category takes — and it collapses the entire legal basis if anyone challenges it.
- · Conversation content encrypted at rest, with a key per person.
- · Data hosted in Brazil.
- · Real deletion, reaching backups and vector memory.
- · None of your data trains a model — neither ours nor anyone else’s.
Integrity report
An unverified commitment is marketing. Every year we publish the numbers that matter:
- Specialists removedand for what reason
- Refunds grantedamount and proportion
- Answers blocked by the guardrailsby category
- Referrals to support linesaggregate volume
Aggregated and identifying no one. The point is to audit the platform, not to watch the people who use it.