How we operate
We would rather have ten people who stay than a thousand who sign up.
Every company says it puts people before growth. Saying it costs nothing, so this page is a list of mechanisms instead: the promise, and the thing in the code that makes it hold.
First things first
The order we do things in
Three questions, and they are answered in this order because answering them out of order is how a product ends up with users it cannot keep.
- 01
She has to arrive on her own.
Either she shows up at the moment it goes wrong, or she is another app that waits to be opened. That is why the phone number is the account. No app, no password, nothing to remember to check.
- 02
Then she has to still be wanted.
Not whether the first week feels clever. Whether someone still wants her on an ordinary Tuesday, once the novelty is gone. Everything else waits behind that question.
- 03
Only then, how it pays for itself.
It is one subscription, priced against what the work is worth. Growth before the first two is a faster way to disappoint more people.
The money
How the money works
Seven promises about money and data. Each one is followed by the mechanism that makes it true, because a promise without one is a sentence anybody could print.
One subscription. That is the whole business model.
$29 a month, billed monthly through Stripe, after a free first week. There is no second revenue line and no tier that unlocks basic decency, because there is only one plan and one price in the code.
The free week ends by charging you, and we say so first.
A card goes on file before the trial starts, and the membership begins on its own when the week is up. You get a text three days out and another the day before, each one naming the date and the price, and each one telling you that cancelling before then means you are never charged.
No advertising, no sponsored placement.
There is no ad-tech in the product: no pixels, no trackers, no data brokers. Nothing she suggests has anyone behind it.
No affiliate links, ever.
Nothing she sends carries a tracking parameter, because they are stripped before the link leaves. Anything starting utm_, plus twenty-three named referral parameters, including Amazon’s. The prompt asks her not to keep them. The stripping is what makes it true instead of a request.
Nothing about you is sold, rented, or used to train anything for anyone else.
Your conversations are processed to run the service you pay for, and that is it. We do not train models on your messages, and the model provider is contractually barred from training on them either. Your mobile number and your consent to be texted are never passed to anyone for marketing.
Leaving is as easy as arriving.
Cancelling takes two messages: say "cancel my subscription", then "yes". Deleting everything takes two as well, and the second is the typed phrase DELETE EVERYTHING rather than a yes. That one is deliberately harder to say by accident, because unlike cancelling it cannot be undone.
Nothing about how much you use her is ever shown to you.
There are no streaks, no scores, no meters and no allowances. Not hidden from the screen: never built. The words themselves are on a banned list the message checker enforces, so the vocabulary cannot reach a text even by accident.
One honest note so the last row is not read too widely. We do record whether messages were delivered, and basic counts that tell us the service is working. That row says shown to you, and that part is exactly true.
Feedback
Telling us we are wrong
Text her and say what is wrong. That is the whole route, and it reaches a person rather than a queue.
There is no form and no ticket number. A message that felt bad to receive is treated here as the most serious kind of bug there is, more serious than one that crashes something, because the whole product is the way it talks to you.
The constraint
ADHD-first is a constraint, not a market
It decides what gets built and, more often, what does not.
Every feature is derived backwards from one symptom and one moment.
Not from a feature list. The working document is organised by symptom, and each entry has to name the moment it fails and how she compensates for it. A feature that cannot fill in those two columns does not get built.
The things we refuse to build are written down.
No streaks, no scores, no urgency marketing. These are standard retention mechanics and they work, on people whose executive function is intact. On the people this is for they produce shame, and shame is the thing that stops someone opening an app again. So they are on a written list that binds the build, not on a list of things we are minded to avoid.
Every outgoing message is checked before it leaves.
That check is code, not a guideline. Every outbound string passes it, templates included, and the sending path is wrapped so that a message which has not declared what it is will not compile. When a message fails the check it does not go out. It falls back to copy that has already passed.
It was not built from the inside.
Nobody here is working from their own diagnosis, and we are not going to borrow anyone else’s to sound like we are. That is exactly why none of this runs on intuition: a feature has to attach to a named symptom at a named moment, a claim is not true here until it has happened on a real phone, and the things we refuse to build are on a written list rather than in someone’s head. Being outside the experience is a reason for the method, not a substitute for it.
Where we are
Where we actually are
The part most companies leave off the page.
There are no users besides the founder. It works, and almost nobody has used it. Both of those are true at the same time, and the second one is the part that usually goes unsaid.
The next thing is ten real people, and one number: how many are still texting her when the novelty is gone.
You can see the same discipline applied feature by feature on how we build, where each row carries what has actually been watched working on a handset.
None of this is a finished argument. What we can promise is that we will keep saying which parts are proven and which parts are still an argument, and that when one of them turns out to be wrong you will read it here rather than work it out for yourself.