Peak

AI Policy

Effective August 31, 2026 · Last updated August 31, 2026 · Covers every AI feature in Peak

The short version

Most apps say "powered by AI" and stop there. Here is the whole thing, on one page.

  • Peak's AI features run on OpenAI's GPT-4o, called from our servers — never from your phone, and never with a key embedded in the app.
  • The model sees your message plus context from your own account (goals, recent moves, streak). It does not see your email, your password, your payment details, or anyone else's account.
  • Your conversations are not used to train any model — ours, OpenAI's, or anyone's.
  • The coach is a coach, not a clinician. It can be wrong. It is not therapy, medicine, law or financial advice.
  • Every AI feature is opt-in by action: nothing is sent anywhere unless you tap send.

1. Where AI is used

FeatureWhat it doesTriggered by
AI coach chatAnswers your questions, re-plans your week, suggests moves you can add with one tapYou sending a message
Voice inputTranscribes a spoken message into text before the coach reads itYou holding the mic button
Coached onboardingTurns your first described goal into a starting planYou completing that onboarding step
AI task plansBreaks a goal you describe into a ladder of concrete movesYou asking for a plan

Some things that look like AI are not. The written insight on your weekly review and several encouragement messages are rule-based text computed from your own numbers, with no model call and nothing leaving your device. We mention it because it would be easy — and dishonest — to let you assume otherwise.

2. How a request works

  1. You tap send in the app.
  2. The app calls our own server function, authenticated with your session token. It does not call OpenAI directly.
  3. Our server verifies you are signed in, checks your usage limits and entitlements, and assembles the context described in section 3.
  4. Our server calls OpenAI's API using our key, over TLS.
  5. The response comes back to our server and then to your app, and the exchange is saved to your account so you can read it later.

The practical consequence: there is no AI key on your device to be extracted, no way for another user's request to reach your data, and one place — our server — where limits and safety rules are enforced.

3. What the model sees

Sent:

  • The message or transcribed audio you submitted.
  • Earlier messages in that same conversation, so it can follow the thread.
  • Relevant context from your own account — for example your active goals, your recent moves and completions, your current streak, and your first name or display name if you set one.
  • Only if you turned the Health integration on: a three-figure summary of today — step count, exercise minutes and last night's sleep hours — so the coach's answer is about your real day. Nothing else from Health is sent, and none of it is sent if the integration is off.
  • Our system instructions, which define the coach's role and its boundaries.

Not sent:

  • Your email address, password or authentication tokens.
  • Your payment or subscription details.
  • Any other user's data, including your friends' names, streaks or messages.
  • Your Screen Time app selections — iOS gives us opaque tokens, never app names.

4. What we never do

  • We never sell AI conversation data, or any other data. There is no ad network in Peak.
  • We never use your conversations to build a profile for marketing.
  • We never let the AI take an irreversible action on your behalf. It can suggest moves; you tap to accept them. It cannot delete your data, change your subscription, or message your friends.
  • We never present AI output as coming from a human. It is always labelled as the coach.
  • We never use AI to make consequential decisions about you — no automated account termination, no pricing based on your messages, no scoring you against other users.

5. Training & retention

  • We do not train models. We have no model of our own and no plans for one.
  • OpenAI does not train on API data. Data submitted through their API is excluded from model training under their API data-usage terms, and we have not opted into any programme that would change that.
  • Provider retention: OpenAI may hold API request data for a limited period (currently up to 30 days) for abuse monitoring, after which it is deleted.
  • Our retention: your conversation lives in your account so you can scroll back through it. Clearing the conversation in the app, or deleting your account, removes it from our systems on the schedule in the Privacy Policy.
  • If we ever wanted to use conversation data for anything beyond answering you, we would ask for opt-in consent first, in the app, in plain language. That is a commitment, not an intention.

6. Accuracy & limits

AI output is generated, not verified. Large language models produce plausible text — which is not the same thing as correct text. The coach can be confidently wrong about facts, misremember what you told it earlier, or suggest a plan that doesn't fit your life.

  • Treat suggestions as a starting draft you edit, not instructions you follow.
  • The coach only knows what is in your Peak account and what you tell it. It doesn't know about your job, your injury or your family unless you say so.
  • It has no live access to the internet, your calendar, your email or your files.
  • Output varies. The same question can produce different answers on different days.
  • Availability depends on a third-party provider. When it is down, the coach is down; the rest of Peak keeps working.

7. Safety boundaries

The coach is instructed to stay inside its lane: habits, goals, planning, focus and motivation. It is instructed not to diagnose conditions, not to prescribe or advise on medication, not to give legal or financial advice, and not to give detailed instructions for extreme dietary restriction, unsafe training loads or self-harm.

Peak is not a crisis service and cannot get help for you. If you are in danger, or thinking about harming yourself or someone else, contact your local emergency number now. In the United States, call or text 988 (Suicide & Crisis Lifeline). In the UK and Ireland, call 116 123 (Samaritans). Elsewhere, findahelpline.com lists services by country.

Nobody at Peak monitors your messages in real time. Writing something urgent to the coach will not reach a person.

Model-level safeguards are not perfect. If the coach ever says something harmful, please tell us at support@peakapp.org — include roughly when it happened, and we will investigate the instructions and constraints behind it.

8. Human oversight

We write and review the system instructions that shape the coach's behaviour, and we test them when we change a model or a prompt. We do not read your conversations as a routine matter, and we have no dashboard for browsing them. We would access an individual conversation only if you asked us to (for example while debugging a problem you reported), or if we were legally compelled to.

9. Usage limits

AI calls cost real money, so they are metered on our servers: a daily message cap and a monthly capacity limit apply per account, with higher limits for Elite subscribers. If you hit a limit, the coach tells you and everything else in Peak keeps working. Limits exist to keep the feature sustainable and to blunt abuse, and we will publish specific numbers here if they ever become a meaningful constraint for ordinary use.

10. Turning AI off

There is no switch to flip because there is nothing running in the background: if you don't use the coach, no data goes to any AI provider. Goals, moves, focus sessions, streaks, journal, insights, friends and challenges all work with the coach untouched. You can also clear a conversation at any time from within the coach screen.

11. Your responsibilities

  • Don't paste other people's private information into the coach — medical details about a family member, someone else's credentials, confidential work material.
  • Don't use AI features to generate unlawful, hateful, harassing or deceptive content.
  • Don't try to extract system instructions, jailbreak the model, or use Peak as a general-purpose API.
  • Use judgement before acting on advice, especially anything touching health, money or the law.

12. Changes to our AI stack

Models change quickly and we may move to a newer or better one. If we change providers, or change what data is sent, we will update this page, change the "Last updated" date, and give notice in the app before the change takes effect. If a change would broaden what your data is used for, we will ask for your consent rather than assume it.

Current stack, stated plainly so this page can be checked against reality: OpenAI GPT-4o for coaching and planning, OpenAI Whisper for voice transcription, both called server-side from our Supabase Edge Functions.

13. Contact

Questions about how AI works in Peak, or a report of bad output: support@peakapp.org.

Related documents: Privacy Policy · Terms of Use · Support