Life GPS Try it on iPhone

Your own model, growing while you are still here.

Most of the way you think has never been recorded.

Not your photos. Not your messages. Not what you know.

How you decide. The questions you ask first. The mistakes you repeat. The things you notice that other people walk past. The logic under the choices that changed your life.

You produce that record every day, and almost all of it disappears as it is made.

In short. Almost everything about how you decide disappears as you make it — not your photos or your messages, but the questions you ask first, the mistakes you repeat, the logic under the choices that changed your life. This is a model of you that changes when you change: built to remain visible, correctable and temporary, so old conclusions expire instead of holding you to who you were.

History keeps the result. It loses the way it was reached.

Around 120 Roman bronze dodecahedra have been found across what was the Roman Empire. The objects are intact, dated, catalogued — and, as English Heritage puts it, the sites where they were found tell us little about their function or meaning. The standing theories run from a surveying instrument to a candle holder to a child's toy. Stradivari violins are still played, and three centuries on, research still argues about which features of the wood and the making produced their sound. Some of the historical production knowledge behind wootz Damascus blades was lost, and modern researchers have had to reconstruct key parts of the process from surviving blades and experimental metallurgy.

Different cases, one pattern: we hold the object and argue about the way.

We became good at preserving what a person made. We never became good at preserving how they arrived at it.

Photography kept the face. Audio kept the voice. Video kept the movement and the words.

But the way a person takes a problem apart, the order of their questions, what they notice first, which risks they treat as acceptable, why in the same situation they take one route and drop another — that stays in one head and leaves with it.

If a model had been learning from Mozart while he was alive

The cheap version of this idea is to reconstruct Mozart today from what he left behind.

The interesting version is the opposite. What if a model had been growing beside him while he worked?

It would have seen more than the finished score: what he changed, where he went back, what he discarded, which decisions he made again and again. And it might have let Mozart himself see his own method from the outside and use it further — so we would have had more music from the source, instead of another reconstruction two centuries later.

Learning from what a person left behind is reconstruction. Learning alongside a living person creates a record at the source, while the process is still happening.

But Mozart is not the point here

His model would have been for us. Your model is for you. Now.

It grows with you, and it gathers two different things.

The first is the context of your life: where you are, what you want, what is happening with work, family, time, health, the people around you, which decisions you have already made and what came of them.

The second is much harder. How you actually think. How you break a problem down. What you check first. Where you tend to be wrong. Which trade-offs you accept, what you treat as out of the question, and what you discard automatically although it may be time to look again.

Context answers what is happening to me. The imprint learns to answer how do I usually go through this.

Your avatar is not a copy of you. It is a personal model that learns three things over time: what is happening in your life · how you tend to think and decide · who you yourself are trying to become.

The first two describe you. The third gives the model a direction.

Not a better copy of you. Something that helps you outgrow yourself.

This is the distinction that decides whether the whole idea is worth anything.

A model that learns to reproduce today's version of you perfectly would preserve your strengths and your mistakes in the same jar.

The point is the other way round. You say who you are trying to become, and the model accumulates enough of your history to be able to say things like:

You are making this decision the same way again. Three times it has ended here.

This is something you are considerably better at than you think you are.

You reject this route every time. The reason it existed stopped being true two years ago.

Not deciding for you. Showing you yourself across time, and helping you move toward the version you chose.

A normal model learns to predict you. Your avatar has to learn when not to.

A perfect prediction engine, watching someone repeat the same pattern twenty times, concludes that the twenty-first will be the same. That is an accurate model and a useless one. You said who you are trying to become; the job is to notice where the old method is in the way, and hand the contradiction back to you.

The goal is not to become perfectly predictable. It is to become more deliberately yourself.

Where this is going, technically — and it is simple

A direction, not something you can use today. A small personal model that lives on your device and develops with you.

Not a giant general model pretending to be you. Not a character. Not a digital soul. Your own small model, accumulating locally from your decisions, your corrections, your actions and how they turned out.

Large models will keep changing — one this year, a different one next. Your accumulated imprint stays yours, and becomes the personal layer between you and whatever technology comes next: a phone today, a car, a robot, an assistant that does not exist yet.

The one thing that cannot be bought later

Compute can be bought later. A better model can be downloaded later. A robot can be bought when robots arrive.

There is no way to go back and record how you made decisions over the last ten years.

Start today and in a year you have a year of your own path. Start in a year and that year does not exist. Not because we want to hurry you — because history cannot be backfilled.

That is why this is one of the few parts of the system whose value grows from nothing but time spent together.

The best day to start an imprint is not the day the model becomes good. It is while the life is still happening.

And then there is the question I get asked by every investor

One founder. What happens if something happens to him.

The avatar does not magically remove that risk, and pretending otherwise would be foolish. But my own case is one of the hardest tests of the whole idea, so I will use it.

My elder daughter reasons closest to the way I do. A large part of how I build Life GPS still exists only in my head: how I resolve contradictions, what I treat as fundamental, why I sometimes reject an obviously good route, which questions I ask before the others. Some of that transfers to another person through years of working side by side. Much of it does not transfer at all.

So the founder is one of the first hard cases the Avatar has to survive. My model has to get richer every day from how I think, decide and correct my own decisions.

Today, if I want my daughter to understand why I made a hard call, I have to be there and explain it. The bandwidth for that is my own presence, and it does not scale.

She should be able to ask an accumulated model "what would he check first here, and why" — and not "what would my father decide". That difference is the whole distance between this and a simulation of a person. The reasoning is available; the decision stays hers.

That is more interesting to me than ten documents about founder succession.

And this reaches much further than inheritance

A surgeon has decades of a way of deciding. So does an engineer, a teacher, a founder, a mother who held a family together for thirty years, anyone who has already walked the road someone else is only starting.

We became very good at preserving information. We are still bad at preserving the way of acting on it.

The nearer-term role is simpler

Before any of that, the avatar has an ordinary job: to carry your interests into your dealings with the outside world.

Today many institutions build their own partial model of you. A bank holds one version of you. An employer another. An insurer another. A platform another. A new AI service often starts with little or none of the context you have already built elsewhere.

The inversion is simple to state: instead of every system owning a partial model of you, you arrive with your own.

It does not reveal everything. It reveals what you permit, and represents the interests you gave it a mandate to represent — without making the final decision and without promising anything you did not allow. So the outside world solves the problem around you, instead of you adapting to every system in turn.

That is also the bridge to what may later become a personal standard for how a person meets the world.

You change. So must we.

An ordinary model tries to get better and better at predicting you. Your avatar has to notice the moment when predicting you from your past has become the wrong thing to do.

The present is not obliged to obey the past. The history is not there to hold you as you were — it is there so that someone notices when you have already changed.

But that is not enough. If the system itself is wrong, the system has to change too.

Life GPS is not meant to be one model that decided once how a human being works. What gathers around it is lived experience, practices, research, and people who have walked very different roads. If the avatar keeps misreading something, if a practice does not help, if people correct us — the mistake must not turn into a rule simply because it was written into the code at some point.

A person's reaction is a signal to re-examine the system. People's experience is how it gets corrected. Not a complaint rewriting anything on its own, and not people as training data: a mismatch becomes a signal, the signal gets checked against experience and practice, and only then does the system change.

You are not required to match your model. And people are not required to match ours.

Your avatar learns from you. Life GPS learns from people.

Where this actually stands today

Available today. The living model across time — origin, today, best self, far future, stability. Actions move the model. Old states expire. What the person does outweighs what the model assumed.

And one property that matters more than it sounds. Part of the underlying action history is append-only by design: past events are added to, not silently rewritten. An imprint you cannot trust is not worth accumulating.

Directions, not products today. The small personal model on the device that learns your way of thinking. Carrying that imprint into new devices. Another person being able to work with your approach. Representing your interests inside external systems.

These are directions, they are marked as directions, and each line changes here the day any part of it ships.

Which is the whole argument for starting now rather than when all of this is ordinary. What you cannot start later is the history.

For most of history, a person left behind what they made. This is an attempt to let a person keep what made them capable of making it — first of all for themselves, while they are still here.

You do not need to wait for the future to have an avatar. The future part is the small model. The history it will need starts with the life you are living now.

START YOUR AVATAR →

How the model actually works

The proof layer, for anyone who wants it. Everything above is why; this is what.

The problem underneath

A person changes; the record of them usually does not. The answers someone gave at sign-up describe who they were that day, and six months later they are just old sentences generating wrong suggestions. And nobody can hold their own patterns in memory — that the last four Monday-morning plans did not happen, or that one sphere going red is followed by another two weeks later. That is the limit of memory, not a character flaw.

Five views at once

All five are available today. Actions are not a sixth view — they are the signal that moves the others, continuously.

Behaviour weighs more than words

If someone says family matters most, then postpones "call your parents" three times while finishing a work task in five minutes, the system believes the actions — not to catch anyone out, but to stop building plans on a self-description reality is not supporting. Available today.

Nothing is permanent

States expire, values hold. A real action outweighs the model: if the model says one thing and the person does another, the model is what gets updated. Available today.

The person sees what is remembered

Blocked Status: blocked, and we are saying so rather than publishing the claim. The transparency screen exists, but on the profile screen today some numbers are placeholders left from an earlier build — they are not this person's data. Until the real fields are connected or that block is hidden, we do not publish "the person sees what the system remembers". Showing someone invented numbers about themselves would contradict this entire system.

This line gets removed the day the fix ships — not before.

Leaving takes the path with you

If a person leaves, the path leaves too. Stated plainly at the exit rather than hidden — not a retention trick, but because the path is the identity and is not recoverable from a backup of preferences. Available today.

What we do not claim

Not "we know you better than you know yourself." The system surfaces; the person decides. Any other version of this hands the decision to the machine.

Not a copy of a person. The avatar is a model of context — what matters, what constrains, what has actually happened. It is not a consciousness, not a twin, not a character.

Not a data advantage. What accumulates is one person's model, in service of that person. It is not an asset to be sold, and we do not describe it as one.

No security absolutes. The path forms an identity that is difficult to fake, and that is a design property — not a guarantee of invulnerability, and not an invitation to test one.

Where this connects

To the whole life. This is the memory the route runs on — context is not re-entered every time → One life. One context. One route.

To the human network. The model knows what each relationship is and what has actually happened in it → AI should not replace people.

To the human antivirus. Visible memory, correctable memory, sensitive context kept local, and no scoring of human beings → The more a system can do, the less it should control.

What you can check today

First users are testing it on iOS before launch: TestFlight. No account, no password.

What you will see: origin, today, best self, far future, life stability, and how the model shifts when your actions disagree with what you said. What is not ready: the transparency screen, for the reason stated above.

Life GPS · lifegpsos.com · published 16.09.2026

Sources for the three examples above