Here is the sequence for Comeback and Desk Reset, the two journeys that use separated passes. On the other journeys, steps 2 and 3 happen together in a single pass — the same information goes in, but one step does both the screening and the writing.
The handover in step 2 is the part that matters. Where the passes are separated, the pass that writes your session never sees your raw answers on their own — it only ever sees them with the screening constraints already attached.
A single pass has to hold two competing jobs at once: be careful, and be useful. Splitting them means the boundaries are decided before the session is written, by a step that has no interest in making the workout impressive — and on some journeys they are checked again afterwards by a step that did not write it.
Internally these passes are given different areas of expertise to reason from — screening, programming, and safety review. They are reasoning steps, not people. There is no panel of experts, and nobody with a professional qualification is involved in producing your program.
This is the most important section on the page, and it is the part we are most confident about.
Some answers are signals that no exercise program is the right response — symptoms that need a person, in the room, with hands and equipment. When those appear, the app stops and points you to professional care instead of writing you a workout. It does not program around them, and it does not soften them.
You are not locked out. If you have already spoken to a professional about it, you can confirm that and continue — and the app then serves a standard, non-personalised template rather than building a program around symptoms it flagged. That choice is recorded.
There are two independent layers doing this, on purpose:
The second layer exists because the first one only protects you when the servers are up. Safety that depends on a network connection is not safety.
Three separate services are involved, and they receive very different things.
| Sent | Never sent |
|---|---|
| Intake answers · current phase and history · comfort and discomfort logs · readiness check-ins · how your finished phases went, as the two 1–5 scores you gave · what you told us about individual exercises (too hard / just right / too easy, liked, skipped) · your additional notes, the restrictions you record for yourself · your daily check-ins as a trend · your age (not your full date of birth) · biological sex, if you provided it · your body weight, if you chose to give it · a generic label rather than your name | Your name · email address · password · payment details · location · the written note and tags from a phase review · anything from Apple Health — steps, sleep, exercise minutes, resting heart rate, VO2 Max, heart rate variability or workouts (the individual readings stay on your iPhone, and the tips beside them are fixed text, not AI) · the Apple Health number made from them, in any form · your mood and feeling words from a Feeling check-in · your saved feeling, Apple Health and overall scores, by day or through the day · your saved daily Apple Health values and sleep stages · anything identifying you as a person |
Under Anthropic's standard API terms, data sent this way is not used to train their models. See Anthropic's Privacy Policy.
When you tap the audio button on an exercise, the voice you hear is generated by ElevenLabs. It receives only the generic description of how that movement is performed — the same words for every person doing that exercise — plus short shared phrases like counted numbers and rest cues.
It never receives your identity, your intake, your comfort scores, or anything about your situation. Your own numbers — your sets, your reps, your comfort ceiling — are assembled on your device from those shared phrases, so your personal figures never leave it.
Audio is stored on our own servers once created and reused after that, so the same guidance is not sent out again. See ElevenLabs' Privacy Policy.
Testers told us the spoken coaching was sometimes hard to follow. So when a new exercise guide is written, we run one automatic check over it, asking a single question: could someone with their eyes closed, in the middle of a set, follow these words exactly — and safely? DeepSeek performs that check.
It receives the exercise name, the prescribed sets and repetitions, and the spoken instruction text. It does not receive your name, your intake answers, your pain logs or your readiness scores. The instruction text was written for you, so it can reflect your situation; we do not send the answers behind it.
The result never reaches you and never changes your program or what you hear — it comes to us, so we can fix guidance that is hard to follow. We use a lower-cost provider for this one routine check to keep the app affordable. Everything that shapes your actual training, including the safety screening that can stop a session, stays with Claude. DeepSeek is based in China; see DeepSeek's Privacy Policy and Section 3b of our Privacy Policy.
Being straight about the limits is part of using it responsibly.
If the program and your body disagree, your body is right. Stop, log it, and speak to a qualified professional.
Used within those limits, this is what the approach is actually for:
HEROMINUTES is a general wellness and fitness guidance app. It is not a medical device. The AI-generated programs, exercise guidance, audio guidance and any insights in the app are for general fitness purposes only. They are not medical advice, and they are not a substitute for assessment, advice or care from a qualified healthcare professional.
Always consult a qualified professional before beginning any exercise programme, particularly if you have an existing injury or condition, are recovering from surgery, are pregnant, or take medication. Never disregard professional advice, or delay seeking it, because of something this app produced.
You use the app at your own risk. See the full Medical Disclaimer and Terms of Use.
If how we use AI changes materially, this page changes with it, and we will tell you in the app.
Questions about any of this: our contact form
Related documents: Privacy Policy · Terms of Use · Medical Disclaimer