Most writing on this question is produced by people who want the answer to be yes. Here is the version with the noes in it, because the noes are more useful.
We should say where we stand first. We used to sell local AI — an assistant on a small machine we offered to place and care for. In 2026 we moved the service into our private cloud on Amazon Web Services, and there is no machine any more. So we have sold it both ways, and we no longer have a side in the local argument. What follows is what we learned.
When it is clearly worth it
When you enjoy building it yourself. Genuinely: if assembling the hardware and tuning the models is the interesting part for you, do that. You will spend less and learn more, and a managed service is the wrong purchase.
When it has to work without the internet. Rarer than people claim, but real: a site with poor connectivity, or work that must continue through an outage. A model running on your own hardware keeps answering when the line is down. Our service does not — it is a cloud service, and without a connection you cannot reach it. Nothing of yours is lost while you are offline, but nothing answers either. If this is your situation, local is the honest recommendation, and it is not us.
When your obligations bar every outside processor. Some confidentiality obligations — ones you did not write and cannot vary — bar handing the material to anyone at all. An obligation like that bars us too: with us, your material would live in our private cloud, handled under a written scope, and for some obligations that is still an outside processor. We would rather say so here than after you have paid — and if your scope is one we cannot meet, we decline it in writing before our letter confirms a year.
When it is clearly not worth it
When you mainly want general knowledge. If your use is drafting, summarising, translating and asking about the world, a hosted assistant at a few dollars a month is better and cheaper, and will stay better. Building a machine to do that would be a mistake — and so would paying us.
When you want it to be somebody else's job. Not about privacy at all. A machine of your own hands you the updates, the backups, and the day it stops working — and the person who built it is usually the only one who can fix it. If what you want is for the thing to work and to be somebody else's problem, local is the wrong direction, whoever you are.
When you have very little material. The value of an assistant of your own comes from an archive. If yours is a few hundred documents, there is not enough for the loading to produce much you could not get by attaching a folder to a hosted chat.
When nobody will own it internally. For a firm, this matters more than the money. If no one is going to decide which contract version is authoritative or what is in scope, the assistant will be built on contradictions and blamed for repeating them — on your own hardware or anywhere else.
The middle, and the test for it
Most people are in the middle: some sensitive material, a decent archive, no wish to run hardware.
The test we would apply is not about privacy or price. It is this: is there a question you want to ask that you currently cannot ask anything?
If you can name three — what did the roof work cost, who has not replied, what changed in the new contract — the material is doing real work, and an assistant built on it is probably worth having. If you cannot name any, the honest recommendation is to wait. Wanting a better assistant in general is not the same as having questions only your own records can answer.
For the people who can name three, the choice used to be a machine; now it is terms. The archive has to be readable by something, and what matters is what that something may do with it. Digital Twin Pro is a private assistant service: one assistant, built on your own material and run entirely by us, in your own encrypted space in our private cloud on Amazon Web Services. Your material is never pooled, never shared, never used to train anyone else's model; it is loaded by hand under a written scope; and when the service ends it is deleted, confirmed in writing.
The money, plainly
$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. 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.
Against a stack of hosted subscriptions that is not a saving in year one, and we have never claimed it is. What you are buying is that your material can be used at all, plus the fact that none of it is your job.
And there is no asset at the end of a year. What ends is the service: your space and everything in it is wiped and deleted, confirmed in writing — and you can take a complete copy of your material at any time, long before the end or on the way out. That is a real reduction in what the year buys compared with a machine you build and keep, and it is better to know that now than to discover it later.
How to find out cheaply
The smaller first step is a written scope of your own material for $1,950 — what can be loaded, what cannot, and a working demonstration built on a sample of your own papers. You keep both whichever way you decide, and if we can see we cannot serve you well we decline and return it.
That exists precisely so this question can be answered with evidence instead of a brochure.
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?