AI that runs
in your building.

A workstation or a small server, sized to the work, with one or more open models installed on it. Your documents are indexed locally so the model can answer from them. There is a simple interface, either a page on your own network or a small desktop application.

The model never leaves the machine, and neither does the question. That is the whole product. Everything else is configuration.

• The documents are confidential. Client files, personnel records, contracts, or anything a regulator has an interest in.
• The bill is predictable. A machine is bought once. It does not meter.
• It has to work without the internet. Job sites, warehouses, and anywhere the connection is not something you would bet a deadline on.
• Somebody has to be able to answer for it. If the answer matters, you should be able to see the model, the version and the documents behind it.

When it does
not make sense.

The page is only useful if this half is as direct as the other.

• If you use AI a handful of times a week. The hardware costs more than the subscription, and it will keep costing you attention.
• If the job needs the strongest model available. For hard reasoning, long multi-step work, or frontier coding, a hosted model is still better value.
• If nobody can look after it. It is a computer on your network. It needs updates, a backup, and somebody who notices when it stops.
• If the answer has to be perfect every time. No model gives you that, local or hosted. This site's whole argument is verification, and it applies here too.

• The machine, specified and configured. Chosen for the models it will run, not for a spec sheet.
• Models installed and pinned to a version, so the answer does not silently change because something updated overnight.
• Your documents indexed, with the index rebuildable from source if it is ever lost.
• An interface somebody will actually use, and a short written guide for the people who will use it.
• An evaluation set. A list of questions with known correct answers, so you can tell whether a change made it better or worse.

Two to four weeks from specification to a machine somebody is using. Most of it is not installation: it is working out which documents matter, how they should be chunked, and what a correct answer looks like.

The hardware is a one-off cost and we will tell you what it is before you commit. The setup is a fixed price. There is no subscription, because there is nothing to subscribe to.

Bring us one job. The right way to start is a single question you want answered from your own files, not a general plan for AI. If that one job works, the rest follows.

Not sure yet? Read the plain-terms explanation first, including the cases where a hosted model is the better purchase.