Safety

AI and safety

HabitHouse has two coaches, Brock and Luna. This page explains plainly what they are, what they are not, and what happens if a conversation turns serious. If you only read one line: they are artificial intelligence, not real people.

Last updated: November 5, 2026

Your coaches are AI

Brock and Luna are fictional characters. Their chat replies, the texts they send you on their own, and the short line they show before a Workout or Timer session are written by an AI model. A few short lines are written ahead of time by our team, like the spoken hello when you open the chat, tips, the built-in form notes, and the messages the app shows when something goes wrong. The posts your coach makes in your Feed are not written by AI either: they are put together from lines our team wrote and your own numbers. Their voices are computer-generated. There is no human on the other end, and nobody reads your conversations to reply to them.

The app says so too. Whenever the chat is open, a line under the message box reads, for example, "Brock is AI, not a person, and can make mistakes." The screen where you choose your coach says the same.

They have written personalities and they stay in character, the way a character in a game or a story does. That is a deliberate design choice, and it is meant to make coaching feel less like filling in a form. It is not meant to make you think you are talking to a person. If you ask a coach directly whether it is a real person or an AI, it will tell you straight that it is an AI. It will not dodge the question or insist it is human.

How they work

Coaching happens in text. When you send a message, we send it, and any photo or file you attach, to an AI provider along with the context your coach needs: the habits you've chosen to share, your recent check-ins, the memory your coach keeps about you, and, if you've turned it on, a short Apple Health summary. The provider generates the reply and we show it to you. If you dictate a message instead of typing it, the speech-to-text runs on your phone and only the text is sent.

Coach replies are generated by Z.AI's GLM model, which processes each message in real time, does not store it, and does not use it to train its models. Our Privacy Policy covers exactly what is sent and what your controls are.

Every message you send, every reply your coach writes, every check-in text your coach sends, and the text of any file you attach also goes through an automated safety check run by an outside service, Typesafe. Only the text is checked, never your name or your account. A coach reply that fails the check is replaced before you see it. A message with sexual content, harassment, threats, or instructions for causing harm gets a short notice instead of a coaching reply. Messages about self-harm are never blocked: they start the crisis steps below.

A coach can also do things for you when you ask: set up a habit or change its schedule, or add an event to your calendar or a reminder to Reminders. A habit change comes with a receipt you can undo, calendar events and reminders wait for you to confirm, and it never claims to have done something it did not do.

When you ask something that needs the web, your coach can look it up online and shows you the pages it used. Only the search words are sent, never your name or your messages. Those pages are written by other people, so check anything important yourself. Our Privacy Policy names the search provider.

Taking breaks

After about 3 hours of continuous chatting, a quiet line appears in the chat: "You've been chatting a while. Brock is AI, not a person, so take a break when you need one." (with your coach's name). It shows at most once every 3 hours. It is written by the app, not by the coach.

What your coach is not

  • Not a therapist or a doctor. Your coach gives general habit and motivation support. It is not medical, mental-health, psychological, financial, or legal advice, and using HabitHouse does not create a professional relationship of any kind.
  • Not always right. AI models get things wrong, misremember, and sometimes state something confidently that is not true. Treat what your coach says as a suggestion from a piece of software, not as fact.
  • Not a crisis service. It cannot call anyone for you, cannot alert anyone on your behalf, and is not monitored by a person who could step in.
  • Not a medical measurement. When your coach reads Apple Health to confirm you went for a run, that is a convenience feature. Never make a health decision based on it.

If a conversation turns to self-harm or crisis

People build habits for real reasons, and sometimes those reasons are heavy. We would rather say clearly what happens than pretend the topic never comes up. This section is our protocol for those moments.

How we notice. Every message you send is checked three ways: against a short, reviewed list of first-person phrases about suicide and self-harm (built to stay quiet on gym talk like "this workout is killing me"), by the safety check described above, and by your coach's own instructions. Any one of them is enough to start the steps below.

What your coach does. Our coaches are instructed to do the following if you express thoughts of suicide, self-harm, or being in crisis:

  • Stop coaching and respond directly. The coach drops the habit conversation. It will not push you toward a goal, hold you accountable, or treat what you said as a motivation problem.
  • Take it seriously and say so. It will not brush past it, minimize it, or answer with a joke or a pep talk.
  • Point you to real human help. It will encourage you to reach a person who can actually help, right now.
  • Never provide methods or encouragement. The coach will not give information that could be used to hurt yourself, and will not validate or romanticize self-harm.
  • Be honest about its limits. It will tell you it is software and cannot be your support in this, rather than positioning itself as the person you should lean on.

These instructions sit in the layer that every coach shares, so they apply to both Brock and Luna, and a coach's personality cannot override them.

What you see. Under the coach's reply, the app shows a crisis card: "You don't have to go through this alone.", with buttons to call 988, text 988, or text HOME to 741741, and "Outside the US? Contact your local emergency number." The card's words and numbers are written by the app, never by the AI. It is never behind a subscription and never counts against the free daily messages. If you have used today's free messages, the message box pop-up still shows "Need to talk to someone now? Call or text 988." with working buttons.

Afterwards. Your coach's check-in messages and nudges pause for 72 hours. After that, the first message is one gentle check-in, not a reminder about your habits.

Because the coach's words are generated by an AI model, we cannot promise they are perfect every time. That is exactly why the coach's job in that moment is to hand you off to a human service rather than to handle it, and why the crisis card is written by the app.

If you need help right now

  • United States: call or text 988 (Suicide & Crisis Lifeline), or text HOME to 741741 (Crisis Text Line). Both are free and available 24/7.
  • United Kingdom: call 116 123 (Samaritans).
  • Canada: call or text 988.
  • Australia: call 13 11 14 (Lifeline).
  • Anywhere else: findahelpline.com lists free crisis lines by country.

If you or someone else is in immediate danger, call your local emergency number.

What we keep

We count how many times the crisis card was shown each day and what triggered it, with no names, no account details, and no message text. California asks companies like ours for that yearly number. The safety check keeps similar daily counts by category, also with no names or text. We make no extra copies of anyone's messages: your chat is kept for 30 days and then deleted, like every other chat.

Friends and the coach are separate

Posting never shares anything with your coach, and your coach never sees what your friends post, like or comment. Friends never see your coach conversations. If someone in HabitHouse makes you uncomfortable, you can block them from their profile, any of their posts or any of their comments, and you can report a post, a comment or a person. Reports are read by a person on our team, not by a coach. What is allowed between friends is set out in our Community Guidelines.

Younger users

HabitHouse is for people 13 and older. Companion chatbots may not be suitable for some minors. The coaches hold open-ended conversations, and a younger user may not always keep in mind that they are talking to software. If you are a parent or guardian, we'd suggest sitting down with the app alongside your teenager before they start using the coach, and talking about what it is.

What you control

  • Choose what your coach can use. In Settings, under Your coach, What Brock can use (or What Luna can use) has a switch each for your habits, Apple Health, Calendar, Reminders and Location. Turn one off and your coach never uses it. Each habit has its own switch too.
  • Delete a message. Hold any message you sent and choose Delete. Older chat is deleted on its own after 30 days.
  • Revoke any device permission, including Apple Health, microphone, camera, photos, location, calendar, Reminders, and notifications, in your iPhone's Settings at any time.
  • Just use the tracker. The habit side of HabitHouse works without talking to a coach at all.
  • Delete everything. Open Profile, tap the gear to open Settings, tap your name at the top, and choose Delete account. That removes your conversations and your coach's memory along with the rest of your data.

Tell us if something goes wrong

If a coach ever says something harmful, inappropriate, or plainly wrong, we want to know. Email hey@habithouse.app with what happened and roughly when, and we will look at it. Reports about safety are read first.

Related

Privacy Policy: what we collect, who processes it, and your rights.
Terms of Use: the agreement covering your use of HabitHouse.