If you are weighing a private assistant of your own against a pile of monthly AI bills, the honest answer to "what does it cost?" is one bill, with an ending. This is the whole 2026 arithmetic for the service that brings the small box your assistant lives in with every assumption shown. No hidden numbers.
The short answer
$8,995 the first year — everything included. $4,995 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 Jetson Orin placed and configured, your archive loaded and the first training run; every year after is the service running. One bill a year: no start fee, no usage charge, no meter, no per-person charge. One home under one roof, or one practice in one office, is the first year — however many of you there are. Several homes, several offices, or staff spread across sites: by conversation, scoped in writing. Not sure which you are? Order the first year; if it needs more, our letter says so before anything is final.
Cost 1: The service (the one bill)
The box arrives fully assembled and tested, built on professional-grade parts, with your assistant already installed and working day one. There is no separate computer, graphics card or build to budget for — the hardware is included in the service.
For technical readers. NVIDIA Jetson Orin NX 16GB: NVIDIA states up to 157 sparse INT8 TOPS (Super Mode), 1024-core Ampere GPU, 16GB LPDDR5, 1TB NVMe, 7W–25W.
Cost 2: The electricity (recurring)
This is the number people overestimate most. The box is built to sip power, not to burn it like a gaming desktop. Leave it on all day and all night and it works out to a few dollars a month. Your own figure depends on your electricity rate and how hard you work it, but even generous assumptions keep it in low single digits.
One bill instead of a stack
The first year is the whole price, and every year after is the service running. The machine stays ours; the model trained on your material is yours alone while the service runs. See how it compares with a stack of rented AI apps →
Required: the service, electricity and internet. Optional: any external AI or API accounts you choose to connect, and the additions below. Because the box does the work itself, there is no per-message bill and no charge per person for supported local workflows.
Cost 3: Optional additions
Your archive loaded before it ships, to the written scope, is already part of the first year, and so are your connections. Beyond that, larger archive loads and media transcription are scoped in writing. None of them are required, and none of them carry a standalone published price.
What you get for that price
Your assistant — yours arrives with a name, and you can change it — comes already installed and set up. Out of the box it handles the everyday work well: drafting, summarising, chatting, listening and speaking, answering from your own papers, and the routines you ask it to keep. All of that happens on the shelf, so supported local workflows can process content without submitting it to an external model provider. And the care is ours while the service runs: updates, health checks, backups — if it breaks, we fix or replace it.
You can also connect an outside model you already pay for for the occasional job that wants them. It is optional, you decide job by job, and you only pay a company on the jobs you send out.
The 3-year total-cost comparison
Here are 36 months of the service against 36 months of renting a stack of AI apps. Every assumption is on the table, so you can swap in your own — and we will not hide the total.
the first year, 3 years
- The first year is the whole price, and every year after is the service running. The machine stays ours; the model trained on your material is yours alone while the service runs.
- Years two and three, if you keep the service: each year after is the service running, at the figure at the top of this page.
- Electricity: a few dollars a month × 36 = about $72 (about 10W on average, left on all day and night, at higher-end U.S. home rates near 30¢ per kilowatt-hour; at the U.S. average nearer 17¢ it is closer to $1.25 a month, so the total here is deliberately not flattering to us)
Rented AI apps, 3 years
- Light user at $29 a month: $1,044
- Moderate user at $50 a month: $1,800
- Heavy user stacking plans at $100 a month: $3,600
The honest arithmetic
We will say the uncomfortable part plainly: on the bill alone, a single $29 chat plan is cheaper than this service, and so is a $100 stack. If the only thing you care about is the smallest possible monthly number, rent one app and stop reading.
- Against a $100-a-month stack: three years of it, and at the end you own nothing and the bills continue. The first year is the whole price, and every year after is the service running — more than that stack's three years, and what it buys is different in kind: a model trained on your own material that is yours alone while the service runs, and the work never left your building. Keep the service after that only if it earns its keep.
- Against a $40–60 household: the service costs more per month than the stack it replaces, and we will not pretend otherwise. What the bill buys that the stack cannot: a model trained on your own material that is yours alone while the service runs, human support with care and repairs on us, no per-person charge, and work that never leaves your building.
- The shape is different: rented apps are a bill that never ends and never adds up to anything. This is one bill for a model that is yours alone while the service runs — and every year you keep paying is a choice you make in writing, thirty days ahead.
So the people best served are the ones already paying for several AI services, and the ones for whom who-holds-your-information is the point. If you pay for one modest plan and privacy is not your concern, the honest case for this service is privacy and care, not undercutting a cheap monthly bill.
Being fair about the limits
A box at home is not a drop-in replacement for the biggest outside services. It is very good at the everyday work: drafting, summarising, speech, answering from your own papers. It is not the tool for the hardest reasoning, or for jobs that need to hold an enormous amount of material in mind at once. Those still go better through a big outside service. That is exactly why you can connect the ones you already use: keep the routine, private, high-volume work at home, and reach out only for the occasional heavyweight job. That mix is usually where the value and the best answers meet.
Frequently asked questions
How much does it really cost per month to run AI at home?
The first year is the whole price, and every year after is the service running — one annual bill, everything included: the Jetson Orin, shipping, setup, your archive loaded, the first training run, and care. Electricity adds a few dollars a month. There is no per-message charge for the work the machine does itself.
What does the first year cost, all in?
The first year is the whole price, and every year after is the service running. The machine stays ours; the model is yours alone while the service runs. Stopping means we collect the machine and delete the model, confirmed in writing.
Does electricity for a home AI box get expensive?
No. The box is built for efficiency, and running it continuously works out to a few dollars a month depending on your rate and how hard you push it. There are no per-message fees, because the work happens on your shelf.
Can it handle any kind of AI job, or are there limits?
It handles drafting, summarising, chat, speech and answering from your own papers very well. The hardest reasoning, and jobs that need to hold an enormous amount of material at once, still go better through a big outside service. You can connect the ones you already use for those, and keep the rest at home.
What happens if I cancel?
The initial term is twelve months, and nothing vests when it ends: the machine stays ours, and the model is yours alone while the service runs. Stop in writing at any time, effective at the end of the paid year; we collect the machine and delete the model, confirmed in writing.
Local AI. Private data. Local training.
$8,995 the first year — everything included. $4,995 each year after. The first year costs more because it contains the Jetson Orin placed and configured, your archive loaded and the first training run; 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?