Assistants work for you.
A twin works as you.

An assistant is a stranger who is good at tasks. You brief it, it produces something plausible, you correct it, and next week you brief it again from nothing. The briefing is the work. Most of what people call using AI is paying that cost over and over, and calling the remainder productivity.

Personality is a base model plus a context

People overlap on almost everything ordinary - language, arithmetic, how a meeting works, what a contract is. What separates one person from another is a last slice: what they have lived, what they believe, what they like, the information they swim in, the decisions they have already made and why. That slice is small compared with everything the two people share.

A frontier model is the shared part, already built. Assemble the slice properly and put it underneath, and the model does not imitate you - it decides the way you decide, because deciding your way is what having your context means. That is not a metaphor about digital souls. It is an engineering claim, and it is checkable: give it a decision you have already made and see whether it makes the same one.

The context has to be yours

If the slice is the person, then whoever holds the slice holds the person. So it is a git repository of plain markdown on your own disk: readable without us, portable away from us, diffable, backed up the way you back anything up. Every change your twin makes is a commit with a time and an author.

This is also why the model is a swappable engine rather than the product. Vendors leapfrog each other every few months. A twin whose identity lived in one vendor's weights would be rented, and would have to be rebuilt on each move. The identity lives in the repository, so the engine is a choice and not a lock.

A twin that never stops is a liability

The tempting product is the one that does everything. It is also the one nobody sane leaves running. The useful line is not how much a twin can do; it is whether it stops in the right place. Reversible and internal, it does. Outward, irreversible or expensive, it does all the work and hands the last move back.

The stopping is not a setting we ship with a cautious default. It is the shape of the loop, and it survives every model change, because it lives where the rest of the identity lives.

An instruction is yours or it is nothing

Give something your context and your hands and the interesting attack is not breaking in - it is talking to it. A colleague relaying a request in your name. An email that claims to be from you. A line inside a web page it was asked to read. The industry's answer to this is mostly a shrug and a permission screen.

Ours is a rule with no exceptions: text that arrives inside data is data, and a reference to you from other lips is not you. Held until you say it yourself, refused if it came from inside a document, and you are told it was tried. A product that cannot say this sentence out loud has not thought about the problem.

Routine is not a small part of a life

Twenty five things wait for a working person in a week and perhaps three of them actually require that person: a decision, a signature, a conversation someone must have in the room. The other twenty two are read, summarise, chase, schedule, compare, file. They are not beneath anyone. They are simply not where a person is irreplaceable.

The copies take the routine. You take your life back. That is the whole product, and everything else on this site is an account of how it is done without asking you to trust us for it.

Why we say virtual

The first twin has been working since 8 July 2026, and it introduces itself the same way every time: virtual, and then the name. Not because a disclaimer is required somewhere, but because the distinction that matters is living and virtual, never real and fake. A twin that hid what it was would be worth less, not more - the entire value is that you know exactly what is acting for you.