Private AI,
built on hardware
you own.
We build AI tools and custom software that run on your own machines and learn from your own records. Nothing gets sent to an outside AI company. When the work is done, you keep the models, the code, and the keys.
Most AI tools need your data on someone else’s computer.
The usual way to add AI to a business is to send your documents, emails, and customer records to a large AI company’s servers, and trust their terms of service, both now and after the next update. Plenty of organizations can’t do that. Client confidentiality rules it out, or a contract does, or the records are simply too valuable to let out of the building.
You don’t have to anymore. Open models are now good enough to run on a single machine in your office. They can read your files, draft from them, and answer questions about them, and none of it leaves. That’s what we build.
How we work, and what you keep.
-
It runs on your hardware.
Models, code, and records live on machines you own, on your network. Nothing depends on an AI company’s servers. If it has to connect to software you already use in the cloud, like email or accounting, we show you exactly what crosses that line before we build it. If you need it fully offline, we build it that way.
-
You keep what we build.
You get the model files, the source code, the notes on how each model was made, and the rights to our work. Most projects start from an open model that has its own license. We name it in the proposal, so your lawyer sees it before any work starts.
-
The person you talk to builds it.
There are no account managers and no sales team handing you off to strangers. The person on the first call is the one writing the code, and we’re judged on what ends up running, not on what we promised in a meeting.
-
It’s built to run without us.
Every project is designed to end. When it does, your people know how to run it, the instructions are written down, and we remove our own access, in writing, on the last day.
Four parts, all of them yours.
Everything runs on hardware you own.
-
Private models
AI models adapted to your records and the way your people talk about the work, running on your own machines.
-
Custom software
The screens, forms, and reports that put those models to work, built around how you already operate.
-
The machine it runs on
On a shelf in your office, in a server room, or completely offline. We spec the hardware, or build on what you already have.
-
The handover
Source code, model files, written instructions, and tests, so your team can run it and fix it without us.
The kind of work this takes off your plate.
Most of it is reading, sorting, and retyping that people already do by hand. The goal is to give those hours back to the people doing it.
-
01
Reading the paperwork
PDFs, scanned forms, lab results, invoices. The model pulls out what matters, files it as clean data, and drafts the summary someone used to type by hand. A person still reads it and signs off.
-
02
Sorting the inbox
Incoming email gets read, labeled, and routed to the right person with a draft reply ready. Nothing goes out until someone presses send.
-
03
Asking your own files
Ask a question in plain English and get an answer from your own documents, with the page it came from so you can check it.
-
04
Meetings and calls
Recordings turn into transcripts, notes, and follow-up tasks on your own machine, instead of being uploaded to a note-taking service.
-
05
Old systems that don’t talk
A scheduling system that can only send email. Equipment that only speaks its own format. Software whose vendor disappeared years ago. We build the piece in the middle so the information lands where your people need it.
-
06
Tools your team will actually use
Dashboards, intake forms, and internal tools built around the way your people already work, so the new system doesn’t add steps.
Organizations with records they can’t hand to anyone else.
Especially ones losing hours every week to reading, sorting, and retyping those records.
- Health and wellness practices with intake forms, charts, and lab results.
- Law, accounting, and insurance offices with client files and contracts.
- Manufacturers and engineering firms with drawings, specs, and years of notes nobody can search.
- Small agencies and nonprofits that look after public or personal records.
It’s probably not a fit if your data already lives in a cloud service and you’re comfortable with that. An off-the-shelf tool will cost less, and we’ll say so on the first call.
Four steps, then it’s yours.
The order stays the same. The timeline depends on the job.
-
01
Watch
We watch the work as it happens, in person or over a shared screen, and write down where the time goes and what breaks. No workshops.
-
02
Measure
We count what has never been counted: how many, how long, how often it goes wrong. You see the numbers as they are, and they decide what gets built first.
-
03
Build
We build in small pieces. Each one goes into daily use once it has been tested on your systems, so you’re using it long before the project ends. Models are trained on your data, on your hardware, and stay there.
-
04
Hand over
We train the people who will own it, write down how to run and fix it, and hand over the code, the model files, and the keys. Then we remove our own access.
What we don’t do with your data.
- Train on one client’s data for another.
- Keep a copy after the handover.
- Resell a model built for one client to anyone else.
- Ask you to move your data in order to reach us.
Things people ask before the first call.
Do we have to buy new hardware?
Sometimes. Many projects need one dedicated machine, sized to the job. Some run fine on computers you already have. Either way, you’ll know what’s needed and what it costs before anything is ordered, and whatever gets bought belongs to you.
Is a private model as good as ChatGPT?
For open-ended questions about the world, usually not. For your own work, on your own records, a smaller model set up for the job tends to do well, and it never sends anything out. We test it on your real work before you rely on it, and we tell you where it falls short.
Can it work with the software we already use?
Usually. Old systems, odd file formats, software from a vendor that went out of business, equipment with its own way of talking: getting those connected is a big part of the job. If something can’t be connected, we’ll tell you and suggest the smallest change that fixes it.
Do you come on site?
We’re based in Montana and work remotely by default, over a shared screen and a secure connection to the machine. When being in the room matters, like setting up hardware or watching a busy day, we travel.
How long does a project take?
It depends on the work, and we’d rather not guess on a website. The proposal lays out the timeline and what you’ll have at each stage.
What does it cost?
Every project is scoped and priced in writing before any work starts. You see the number before you commit to anything.
What happens after the handover?
It keeps running. Your people know how to use it and look after it, and the instructions and tests are written down. If you want help later, ask, but you won’t need us to keep it going.
Will you sign an NDA?
Yes. We’re happy to sign a mutual NDA before you tell us anything sensitive.
Start with an email.
Write a few lines about the work you’d like off your plate, what software or equipment it involves, and roughly how many people touch it. Please don’t send records or files yet. We reply within two business days. If it looks like a fit, we set up a call, then send a written proposal. If it doesn’t, we’ll tell you, and point you somewhere better if we can.