iOS, in development

The plan gets written for your cat. Not picked off a shelf.

Lapka teaches cats tricks, and helps you undo the habits you'd rather not live with. You describe your cat in your own words, and a model writes the plan out of what you said. The second half of that idea — a set of questions written for your cat and nobody else's — is built and not yet reachable, and the band about it below says exactly where that stands.

Lapka isn't on the App Store yet.

It's being built in the open, iOS first. There's nothing to install today, and there's no form here asking for your email — there's just an address you can write to if you want to know when it's out.

Everything described below is either something you can see running in a screenshot, or something with a line under it saying it isn't yet. The pictures were taken in August and the app has moved since — where one of them shows a screen that has been rewritten or deleted, its caption says so.

Where it actually is, in seven lines →

The Lapka programme screen on an iPhone, as it was in August 2026 — this ladder has since been deleted. A card headed Today reads 'Sits calmly in front of you', two minutes, with a button labelled 'Tell us how it went'. Below it, 'The whole path — step 1 of 10' and a vertical ladder of ten steps, all of them readable, none locked. — The programme screen, captured on a simulator in August. Every step is visible from day one and none of them is locked — that part is still true. **The ladder in the picture is not.** These ten steps were hand-written, the same for every cat, and on 4 September they were deleted along with both paths that still drew them. A plan is written by a model now, for the cat in the intake, and it runs twelve rungs rather than always these ten. This screenshot has not been re-shot yet.

One sign is never enough to go on

A quiz-shaped app narrows a problem with a single question and hands you the guide that matches. Take a problem where somebody has actually measured how well that works — a cat weeing somewhere that isn't the tray.

In a study of 245 cats, each of the standard signs was scored against the diagnosis for the 132 of them who were weeing outside the tray. "On a vertical surface" — exactly the sort of thing a quiz branches on — picks out only 66.7% of the cats who are marking, so a third of them come out on the wrong side of that one question. The best single sign is posture, and it is the one owners least often get to see: in an Italian survey of 3,106 cats, 34.2% of owners had never watched their cat's posture at all.

"Individual cardinal signs were generally not good predictors of diagnosis."

Barcelos, McPeake, Affenzeller & Mills (2018), Frontiers in Veterinary Science, 245 cats. The 34.2% is Tateo et al. (2023), Animals, 3,106 cats. Both are veterinary papers about diagnosis, which is not what Lapka does — we cite them as evidence about questions.

That is our reading of it, and it is the whole argument for building this: an app that decides from one answer is making a call it has not earned. So Lapka doesn't try to. It uses what you have already told it to work out what to ask next, and it asks about your cat rather than about cats. Half of that is running: the plan itself is written from everything you said, not chosen from a shelf. The other half — using your answers to work out what to ask you next — is written and not yet reachable, and the band below says exactly where it stands.

It never tells you what's wrong with your cat. That is a vet's job. Lapka's job is what to do in your living room this evening.

The intake · 1

A name, then nine questions — twelve if something's wrong

The name comes first and on its own, because everything after it is about them by name. Then nine questions everybody gets, and each one changes something downstream:

How old is Mochi?
Is this your first cat?
When did Mochi come home?
Which breed is Mochi? — a list, and "mixed" is first
What does Mochi look like? — only if you said mixed, or didn't say
Where does Mochi live? — a flat or a house, and whether they go outside
What would you like most?
What's bothering you most? — pick up to three
Who else lives with Mochi? — on two feet or four
How much time can you give Mochi on a normal day? — "Five good minutes beat twenty distracted ones."

Ten rows, nine questions: the two in the middle are one question that knows when to stop. A pedigree name already fixes the coat, the frame and the face, so a Persian is never asked whether it is flat-faced. "Mixed" fixes nothing, which is the one answer that has to be followed up.

The answers to "what would you like most" are: put a stop to a bad habit · teach some tricks · grow closer to my cat · understand cats better. Only one of those is a complaint, and "Nothing, all is well" is a real answer to the next question. We don't assume anything is wrong.

Three more arrive only for somebody who said they'd like to put a stop to a bad habit: when it started, whether your cat is a girl or a boy and whether they're neutered, and what else has changed at home. Until 12 August 2026 those three were quiet rows on the confirmation screen that no question had ever filled, which is how an owner ended up staring at a dead button. They are screens now.

The intake · 2

Then a paragraph, in your own words

A box and a nudge — like you'd tell a friend, messy is fine. There's a microphone beside it if saying it out loud is easier.

Nothing reads it on the way past. It is kept exactly as you typed it, shown back to you exactly as you typed it, and handed to whatever writes your plan untouched. Nothing summarises it, nothing pulls fields out of it, nothing quotes it back at you in a tidier voice. They're your words and they stay yours.

One exception, and it is a small one: the three urgent-sign checks below run a short list of hand-written phrases over this text — the typed part of it, never the words we wrote down from a recording. That is a keyword match, not a reading — it can open a question and nothing else.

The Lapka intake screen headed 'What else should we know about Mochi?'. A glass card labelled 'In your own words' holds a typed paragraph about a cat who follows his owner round the flat, a wish to teach a high five, and scratching the arm of the sofa. Beneath it, the hint 'Like you'd tell a friend — messy is fine' and a microphone row reading 'Or say it out loud'. — The paragraph screen, as it runs today. The voice note records, plays back and deletes — and since 12 August it is **turned into text** by a transcription service. That used to happen after you signed up; since 23 August it happens right here, before there is an account and long before there is anything to pay, because sign-up is the last step and the words were arriving after both screens that were meant to show them. This screen shows you the words; the next one is where you can fix them. The recording isn't kept once the words are back; the privacy notice's §7 is entirely about that one round trip.

The intake · 3

Everything back on one screen, before anything gets built

Every answer in a list you can change, with your own paragraph quoted underneath it. A few rows are marked in the accent because they are still blank, and those are the ones a plan can't be built without.

It is the last screen before the price, and that is deliberate. You should see what we're reasoning from before you decide whether it's worth anything.

A Lapka screen headed 'So — this is Mochi?', as it was in August 2026, with the subtitle 'Change anything that's off. This is what the plan gets built from.' A card of editable rows: What you'd like most — Teach some tricks; When it started; Age — Adult; Sex; What else changed; Who's at home — Other grown-ups. Three rows are set in teal because they are still unanswered, and the action bar reads 'The highlighted rows still need you' above a disabled button, 'Yes, that's us'. — "So — this?" in August. It has thirteen rows now rather than six, every one of them is asked as its own screen so they arrive already answered, and anything you said out loud is quoted underneath with a way to fix what we heard. The accent and the disabled button are still there for a row you skipped — but the version in this picture, with three medical rows under a goal of "teach some tricks", is a state the app stopped being able to produce on 12 August.

The intake · 4

Then the questions that didn't exist until your cat did

This is the part a fixed app can't do. Once there's a cat, the server writes a second, shorter set of questions — five to seven of them — for that cat. If it's the sofa: what exactly is being scratched, how tall the post is, whether it wobbles. Nobody could have written those in advance, because they are about your post, in your hall.

And here is where that stands, and since August it has gone backwards. The server writes the set, a schema keeps it to five to seven questions of types the app can already draw, and since 11 August there has been a screen that draws them, translated into all three languages. Nothing takes you to it. The only way in is a developer shortcut that is compiled out of every release build, so after the price the app goes straight to writing the plan — and every programme written so far has been written with this second set empty. The door is the next thing being built. Until it is, this band describes a screen no owner has ever seen, and saying so is cheaper than the alternative.

One thing to do today. Then you tell us how it went.

The plan is a ladder of small steps and you can see all of it from the first day. Nothing is locked. Today is one rung and it takes minutes.

Afterwards you file a short report — four questions, no typing. The first one is a gate: a session that didn't happen ends it there and changes nothing at all.

Three calm sessions in a row open the next step. An unsettled one doesn't undo them — that is an ordinary session, and a rule that wiped out a run over it would only teach people to tap "calm". One that was too much empties the run and starts it again, on the same step. For a habit you're undoing the bar is different: a day with no incidents is what clears it.

Where you are on the ladder is worked out from your reports every single time it is read, and never kept as a score. So if you go back and correct yesterday's report, everything after it is simply right — there is no counter to break.

Why a ladder, and why the steps are so small

The best evidence we found for this isn't about cats at all. A 2026 meta-analysis in Health Psychology Review pooled 39 randomised trials of graded tasks in human physical activity, N = 4,190: g = 0.61 (95% CI 0.42–0.81, p < .001), and the number of graded increments on its own explained 47% of the variance in outcome (R² = 0.47, β = 0.68, p < .001).

The caveats belong in the same breath. Heterogeneity was high (I² = 83%), the authors rate the certainty of the evidence low, and graded tasks were always delivered alongside goal setting and action planning — the three travel together, so g = 0.61 cannot be assigned to grading alone. There are no cats in it. We are borrowing a mechanic, not a species.

Where we stop, and hand over

Lapka is a behaviour trainer. It is not a vet, it doesn't examine your cat, and it will never tell you what's wrong with them.

There is one thing it does with health, and it is a handover rather than help. Three signs — breathing, back legs, eating — each have one plain question attached. Only your answer to that question can stop the app; nothing fires because of a word we spotted. When it does stop, it says your cat needs a vet and roughly how soon, gives you the sentence to say on the phone, and offers to find one near you. Then it stops — in its own words, "we can't examine your cat and we can't tell you what this is." It runs during the questions, before there is an account and long before there is a price.

And its limits, said plainly, because a safety claim that flatters itself is worse than none. All three are opened only by matching English phrasing in what you typed, in an app that also ships in Ukrainian and Spanish. There used to be a fourth sign, for urinating, and it was the only door here that opened from a tick rather than from words — it was removed on 12 August 2026, and it took the layer's only language-independent door with it. Somebody who records a voice note instead of typing opens nothing at all. Since 23 August the words come back before sign-up rather than after it, so the timing is no longer the reason — the reason is that the recording and what we wrote down from it are kept as a separate thing from the paragraph, and the checks only ever read what was typed. An owner who says all of it out loud gets no checks, and neither does one who writes nothing. This is best-effort recall, not coverage. It is not a safety net, and it will miss things.

There is no picture of this here any more. The one that was is the urinary check, which was removed on 12 August 2026, and a photograph of a screen that no longer exists is the worst thing a page like this can carry. A new one, of a check that still runs, goes here when it is shot.

Things Lapka won't do

Some of these cost a feature people expect. They aren't going to change.

No punishment, and no blame

Nothing in a Lapka plan will ask you to startle a cat, spray one, or tell one off. And nothing in it treats you as the problem. People usually arrive having already tried hard.

No streaks, and nothing locked

In a fitness-app experiment, people carried on about 66% of the time with a streak intact against about 58% once it had broken — and in that condition the app itself had caused the break. So there is no chain here to lose and nothing behind a padlock. Anything we count of your own effort only ever goes up.

Silverman & Barasch (2023), Journal of Consumer Research 49(6).

No course on reading cat body language

It sounds like the obvious feature. In a randomised trial with 368 adults, a 2.5-minute training video gave three or four points on the obvious signals and made recognition of the subtle negative ones 18.8 points worse. So we don't teach it as a subject. You pick up one signal at a time, on your own cat, attached to the session it turned up in.

Henning, Nielsen, Hazel & Atkinson (2025), Frontiers in Ethology. The sample was more expert than the general public, which makes those numbers worse rather than better.

A plan never mentions your cat's breed

We do ask it now — this card used to say it was the first thing people expect to be asked and we weren't going to. The name is doing two jobs and only one of them was a bad bet: it is who your cat is, which is worth showing back to you, and it is a shortcut to three things a plan genuinely uses — coat, frame, face. A Persian is flat-faced and a Maine Coon is huge, and inferring that is not a guess. Say "mixed" and we ask what your cat actually looks like instead, because "mixed" fixes nothing.

The refusal itself is unchanged, and it is a constraint rather than a promise. The breed can decide which piece of sourced knowledge applies to your cat; it can never be printed into the sentence you read. Breed selects, breed never speaks — the database rejects the write, so it is not a rule anybody has to remember. A plan tells you what follows from the breed, and never the breed.

Half of this card was rewritten on 12 August 2026.

No canned intelligence

If the model can't answer something well yet, Lapka says so rather than producing something with the shape of an answer. That rule is why this page reads the way it does.

No ratings, no reviews, no happy-cat count

Two people have used it — an Android build went to two friends on 31 August — and neither got as far as a plan for their own cat, so there is still nothing here from anybody who has. A made-up case study is the easiest thing on earth to write, which is exactly what it's worth.

What it will cost, when there's something to buy

$2.99 a month, or $24.99 a year — which works out at $2.08 a month.

Three days free to start — on the monthly plan, and only there; a year bought outright begins paid, and it is one free trial per person rather than one per plan. The free days are the paid plan with the billing held back, not a smaller version of it, and there is no free tier: without a subscription there is nothing to use. Billing goes through whichever store sold it — the App Store on an iPhone, Google Play on Android — and cancelling happens there too.

There is no web checkout at a different price, and there won't be. Billing is the most complained-about thing in this whole category, and a good share of those complaints are about exactly that.

Nothing is being charged today, because there is nothing yet to buy.

The Lapka paywall as it was in August 2026; the screen has been rewritten since. Heading: 'We're ready to build Mochi's programme', subtitle 'Here's where we start — tonight.' A card headed 'What we're going on' lists what you'd like to work on (teach some tricks), how old your cat is (adult) and who else lives with you. Two plans: '3 days free — then $2.99 a month, cancel any time', selected, and '$24.99 a year, −30%, works out at $2.08 a month'. The button reads 'Start my 3 free days' above 'Renews automatically. Cancel any time in your App Store settings.' — The price screen as it was in August — the words on it have changed since. The plumbing behind it is real now: the app asks the store for its prices and the server verifies what comes back. What it has not got is anything to sell — the Paid Apps Agreement isn't signed, so Apple hands the app no products and the screen says so out loud. Nobody has ever been charged.

Where this actually is, today

The seven lines the hero promised. Four of them are good news, which is two more than on 9 August. Three were rewritten on 12 August and five more on 5 September, because they had stopped being true — and a page that argues for not overstating things has to survive its own good weeks as well as its bad ones.

Not on the App Store — but there is a build now.

Both things this row used to say were missing have arrived. The Apple Developer account exists, the App Store Connect record exists, and on 4 September the first build was uploaded to TestFlight and accepted. What is still missing is the Paid Apps Agreement: until that is signed Apple hands the app no products at all, so the price screen says it can't reach the store and it is telling the truth. There is still nothing here to install.
The app runs, on a phone, end to end — except at the price.

The questions, the paragraph, the urgent-sign checks, the ladder, the session reports. It remembers all of it between launches in a small database on the device, and once there is an account the questions, the paragraph, the ladder and the reports go up to the server. What you answered on an urgent-sign check does not — that stays on the phone, and §6 of the privacy notice says exactly what the server does learn about one. The one place it stops is the price: the Paid Apps Agreement isn't signed, so Apple hands the app no products. Every screenshot on this page came off a simulator running it, in August — three of them now show screens that have been rewritten or deleted, and each of those says so underneath.
The plan is written by a model now, and the writing isn't good enough yet.

This row said "written by hand" until 10 August, and the hand-written ladder was the honest description. The server generates the programme for the cat in front of it now — but read the second half. While there was still a hand-written ladder to hold it next to, a generated one had got easier partway up instead of harder, and had left the furniture covered from the first rung to the last, which makes "no scratching for a week" a test you cannot fail. The hand-written one was deleted on 4 September, so there is nothing left to compare against — only a list of things we keep finding wrong with the writing. It is the real thing and it is not yet the good thing.
The models are switched on.

Also new since this page was written: the keys are set, real programmes have been written, and voice notes are turned into text by a real service rather than a stand-in. Nothing on this page is a screenshot of that — the pictures are older than it is. No model has yet written a plan for anybody's cat but the builder's.
There is a server, and it is running.

A rented machine in Finland, answering every route the app needs against a real database, over HTTPS. It went up on 12 August; before that it ran on a laptop. The app talks to it before you have an account in exactly one case, and it has since 23 August: a voice note recorded on the paragraph screen goes up to be written down, because sign-up is the last step of the intake and the words were otherwise arriving after both screens meant to show them. Everything else waits for the account, and the account comes after all the questions — so somebody who types their cat's story and then walks away has still told a server nothing. Somebody who said it out loud has.
Two people have used it. Nobody has used it for their cat.

On 31 August an Android build went to two friends and they sent back five things — one of them could not get past the sign-in screen. That is all of it. No ratings, no reviews, no stories, no numbers about how well it works, because neither of them ever reached a plan: the only three accounts that can get past the price are ours.
English, Ukrainian and Spanish, from the first release.

This page is English only. A legal page put through a machine translator is worse than one that isn't there yet, so the other two wait for a person.

Want to know when it's out?

There is still nothing to sign up to, and that is deliberate rather than an oversight. A form on a page like this has to post somewhere, which means a third party holding a list of email addresses — and this site also ships a privacy notice, which would then have to say how long that list is kept and who "we" is as a legal entity. Neither is decided. Collecting something we can't yet describe is worse than not collecting it, and a list you were added to is a different thing from a message you chose to send.

Write to lapka.admin@gmail.com and the person building it will answer you. There's no list, no automation and no newsletter behind it — it is an email to a person, and it is the same address on all three pages.

The address is printed rather than hidden behind the link, because a mail link does nothing at all on a machine with no mail app set up. Added 12 August 2026; the privacy notice's §9 says what happens to what you send there, and its §10 is the other thing this address is for — asking us to delete what we hold.