Technology learned how to capture human attention. We are building the opposite layer.
In short
As AI becomes more capable, safety is not only a question of what a model can do. It is also a question of what stays under human control.
Life GPS is built inside eighteen constraints that were written before the code. They say what the system may not do — even when doing it would improve every metric a normal product optimises. Protection can override progress. The model of a person is designed to be visible and correctable by them. Sensitive context is local-first: only explicitly permitted fields leave the device. People are never rated. And the system does not need to keep a person inside it in order to have succeeded.
The more a system can do for a person, the less it should control that person.
This is not a privacy page. It is the constitution the other three systems run inside.
The problem this addresses
An entire generation of products was optimised for time spent. The mechanics are well known: infinite feeds, streaks you can lose, counters of things left unread, notifications engineered around the fear of missing out. They work — that is the problem.
The pressure is now legal as well as ethical. In March 2026 a Los Angeles jury found Meta and Google negligent over platform design in a case about harms to a young user; the disputed features included infinite scroll and autoplay. In August 2026 state attorneys general announced a multistate settlement with Meta of up to $17.1 billion, alongside binding product-design and child-safety requirements. California describes that settlement as proposed and subject to court approval. The fight has moved from what content was shown to how the product itself was built.
What this does not prove. Not that these mechanisms are universally unlawful, and not that Life GPS is clinically or legally safer. It establishes that product architecture is now a subject of liability — which is the environment any new system is built in.
And there is a version of the same risk that has nothing to do with feeds: a system that holds a person's whole life could steer that life without ever looking like it was steering. Guarding against that is what most of this system is for.
What Life GPS protects
Attention · time · choice · mental state · relationships · a person from their own overload · a person from the pressure of their environment · a person from consumption that does not serve them · and a person from the pressure of this system itself.
The last one is the one nobody else writes down.
What that means in the product
Protection can override the goal
Three days without progress, energy at the floor, a sphere in the red — and the system does not push harder. It lowers the load. Silence is a valid answer. A short session is a good session.
In the beta.
No streaks to lose
There is a daily check-in — a mark a person makes on their own route. It is their route: they set the tasks, they set the reminder time. The labels state facts — "days in a row", "days total" — not an achievement that can be taken away.
(This is not the same thing as Daily Guidance — a morning task and an evening check-in generated by the system, which is built and switched off. The check-in below is live; the guidance around it is not.)
The forbidden thing was never the mark. It was the frame of loss: don't break it, don't lose it, keep it alive. Those phrasings are named and banned in the canon, and they are out of the product as of 16.09.2026.
In the beta. (We do not claim "no gamification" as an absolute — one area is still under review, and we will say so until it is checked.)
No account, no feed, no posts
A person opens the app and starts. No email, no password, no profile to fill in, no audience, nothing to scroll. This is not a missing feature — it is a design result: it turned out not to be needed.
In the beta.
Sensitive context stays local
Sensitive life context is held on the device. Only explicitly permitted fields leave it.
We do not claim "we collect nothing" — an absolute like that is broken by a single line of telemetry, and one broken absolute discredits everything else on this page. The precise list of what leaves the device belongs here and is being prepared.
Crisis is handled offline
A crisis signal stops the route hard and surfaces help. It works without a connection, because that is exactly when a connection cannot be assumed.
In the beta.
Protection is protocols, not a switch
Different situations need different protections, and they are defined separately rather than bundled into one setting a person has to find.
In the beta.
People are never rated
No scores on human beings, no public ratings, no quality numbers, no leaderboards. This is a constitutional rule in the network canon: a violation of it cancels the feature, not the rule.
In the beta.
The person can always refuse, change, or leave
A refusal is a normal outcome, not a failure state. A person can change what they said, contradict the model, or leave with their path.
In the beta.
The internet, inside a route instead of around it
This is where the antivirus stops being defensive and starts being the product.
The internet should not be somewhere you disappear into. It should become a resource inside your life.
- scrolling → filtering
- a feed → a route
- advertising → relevance
- attention → intent
- practices, experts and support → appearing at the moment of real need, not competing for a slot in a feed
- content → becoming an action inside a life, instead of something consumed without end
We are not against business. We want a better moment between a business and a person: fewer random impressions, more relevance when a real need exists. Quantity belongs in creation. Quality belongs in consumption.
Favourites and support inside the route are in the beta. Partners and experts entering the route at the moment of need is a direction we are building toward.
What we publish and what we do not
We publish what the system protects, what it considers, what the person sees, and what result it is designed to produce. We do not publish the internal weighting, thresholds, detection logic or matching mechanics that make it work.
Architecture and behaviour, yes. The recipe for the engine, no.
The eighteen constraints
They were fixed in May 2026, before the code that implements them, and every new feature, field, screen and role is checked against them. They cover: the human · the snapshot · the influence graph · personal constraints · protection · the route · the response plan · validators · the avatar · the network · environment · world · signal extraction · memory and decay · no direct shortcuts from data to advice · explainability · the guardian itself · and clarification before assumption.
Full text: /principles.
What this does not claim. That every line of code is perfect, or that constraints written down are constraints never violated. It claims that the constraints exist, predate the code, are public, and are checkable — including by people who would like to catch us.
Where this connects
To the whole life. Every step the route produces passes through these limits before it reaches a person → One life. One context. One route.
To the human network. No ratings, no marketplace, no pressure to respond; safety routing overrides everything → AI should not replace people.
To the living avatar. Visible memory, correctable memory, decay, and the right to contradict the model → A person is not a profile.
What you can check today
The iOS beta is open: TestFlight. No account, no password — that claim is the first thing you can verify, in the first fifteen seconds.
Then look for what is not there: no feed, no streak to lose, no unread counters, no way to be rated by anyone.
Life GPS · lifegpsos.com · published 16.09.2026
Published 2026-09-16 · Updated 2026-09-16 · Every claim on this page carries a status