More people are asking the same question: can I have a capable AI assistant at home — private, on a machine of my own? In 2026 the honest answer comes in two halves. Yes, you can run one at home. And no, the machine was never the part that made it private. This guide gives both halves plainly — and then says what we sell instead, because we used to sell the machine, and we stopped.
What you can genuinely run at home
The small open models got good. On a capable computer — a recent Mac, a desktop with a serious graphics card — they draft, summarise, transcribe and answer questions over documents you have indexed yourself, and for that work no request leaves the room. That is real, and it would be dishonest of us to pretend otherwise.
What the enthusiastic guides skip is the second column. A home build costs a weekend to stand up and a hobby's worth of care to keep standing: models to update, indexes to rebuild, backups to remember, and the quiet Sunday when something breaks. The models that fit on a home machine are also not the strongest models, and on the hardest work — long reasoning, a decade of material held in mind at once — the gap shows. If you enjoy that kind of project, build, sincerely; some of the happiest people in this field run their own.
What the box does not settle
Here is the part we most want you to take from this page, because we learned it the slow way: privacy is a property of design, not of location. A machine can run in your house and still quietly report back about you. A service can run far away and be built so that nothing moves without your say-so, so that your material builds nobody's product, and so that what happens is written down where you can check it.
The questions that decide it are the same in both cases: what leaves without your approval, who can read what, whether what you type is used to build someone else's model, what is kept, and what is deleted when you go. A box on a shelf answers none of those by itself — and it hands you the door lock, the Wi-Fi, the updates and the backups as your jobs.
What we sell now, and why it changed
We spent our first chapter building the service around a small machine of its own, so we say this with no contempt for the idea. We stopped because the promise outgrew the box: the strongest commercial models — the ones that make an assistant genuinely worth having — could not run on the hardware we offered to place, and we were not willing to sell second-best thinking to keep a shape we were fond of.
What we run instead is one assistant, yours alone, in an encrypted space of your own in our private cloud on Amazon Web Services. Your archive is prepared and loaded by hand under a written scope, and your assistant is built on it, so answers come from your own records, with citations. Your material is never pooled, never shared, never used to train anyone else's model. The models it thinks with are reached through Amazon Bedrock, inside the same boundary as everything else, over private networking rather than the public internet, and they are part of the fee — chosen by us for each kind of work, with no meter running. Nothing is sent or booked without your yes. And because there is no machine to build or ship, it is ready in days.
For the reader who wants the box on principle: dedicated hardware is deferred, not abandoned. The architecture keeps a clean place for it, and we will add it when it is warranted. We promise neither a date nor that it will happen — the door is open, and that is all we can say truthfully.
What using it is like
There is no setup that is yours, and you will not touch anything that looks like a command line. Your assistant has a name from the first day, and you can change it. You write to it from the apps you already use — Telegram, or plain email — and it answers from your own records, showing where each answer came from. Then the work begins: questions about your papers, summaries, drafts, and follow-ups it brings back for your yes. It is only ever as good as what you hand us, which is why the loading comes first, under a scope you hold in writing.
What it costs
$9,995 the first year — everything included. $5,995 each year after. Before it, if you like, the Archive Assessment: $1,950, credited in full against the first year. The first year costs more because it contains the preparation — your archive prepared and loaded by hand under a written scope, and your assistant built on it — as well as the first year of the service; every year after is the service running. One bill a year: no start fee, no usage charge, no per-person charge, and using it more does not move it.
The bottom line
Can you run a private AI assistant at home in 2026? Yes — and if the building of it is the part you would enjoy, you should. But if what you actually wanted was the privacy and the help, without the weekend or the Sundays after it, the machine was never the point. One assistant, yours alone, built on your own material and run entirely by us, is now the most practical way to have one.
Rent an app, build it yourself, or have it run for you →
Your own AI. Your own material. One assistant, yours alone.
$9,995 the first year — everything included. $5,995 each year after. The first year costs more because it contains the preparation — your archive prepared and loaded by hand under a written scope, and your assistant built on it — as well as the first year of the service; every year after is the service running.
Or add the Archive Assessment first — $1,950, credited in full
You order and pay at checkout; we write back within days — a decline returns every dollar, and until our letter confirms the year you may withdraw in writing. From that letter the year is final, and it arrives on the date the letter names. Prefer to write to us first?