What you buy, what it costs, and the rules.

If you build or move things and some of your work waits on one or two experienced people, probably yes. We're starting with machine shops because a quote desk is easy to measure. The AI Readiness Check will give you a straight answer.

No. It's built for every desk, from the first lead to the final invoice: lead gen and sales, operations on the floor and at the dock, HR and finance. One model, an agent for each desk, and a named person decides at each. In HR it never screens, ranks or scores anyone, and never reads a job application. In finance it never pays, posts an entry or sends an invoice, and it flags any request it sees to change a supplier's bank details as unverified, for a person to confirm by calling a number already on file. In a smaller company one person often covers several desks; Si drafts for the work, not the job title. The Audit picks where to start.

No. It reads, checks, finds and drafts. A named person decides anything public, financial or legal, and anything that moves a machine. On a quote desk, your estimator still sets every hour and every price.

No. It does the paperwork around them: the job packet at the right revision, the first-article form set up for your inspector, the shift handover, the shipping paperwork. It drafts from the records you already keep, and a person checks each one. It never starts, stops or moves a machine, never overrides a safety check, and never fills in a measurement or passes a part. Your people make and move the things.

Today, nothing directly: it starts from the exports your systems already make (CSV or Excel). Read-only connectors are planned for the systems you already run, such as email and files, your ERP, accounting, CRM, shipping and payroll (the list). None is built yet. It never changes your records.

No. A server is in your monthly price: we own it, insure it and look after it in your building, and it's yours after month 36. Or your model can run in a private cloud in your own account, for $250 a month less. Some customers' terms rule out the cloud; month 1 checks yours.

Open-weight NVIDIA Nemotron models today. Every model that writes answers on our rack is an NVIDIA Nemotron: Super 120B, Lightning 30B, Nano 4B and Nano 12B VL. The Audit picks the base for your workflow, and the version is pinned until a new one passes your own test set. The open models that fit on one server trail the best rented models (where rented superintelligence is better).

It makes your model answer the way your company works: your formats, your terms, your next steps. It doesn't make it smarter than your people, and it isn't how names get on answers; your records do that. In our one fine-tune so far, our own side-by-side check found it answered in that company's working format. It has no test score yet. Our first blind test, on the quote desk, comes next; it tests the quote desk, not this fine-tune (what we measured).

No. Your data, and the fine-tune built from it, are never used for any other company, including any company we buy later. Fine-tunes are never merged or pooled.

They can stop at any time and review what they said. What each person can take back, and what your company keeps, is set in the consent terms they sign, which employment counsel is reviewing. One limit is already clear: a trained model can't forget one person's part on request; it comes out only when the model is retrained without it, and answers already given aren't recalled. We'll publish the exact rule here before our first Build.

That's being settled in our terms now. You keep your data and a full export of everything your Si keeps. Your fine-tune is a small file that only runs on top of the open model it was trained on, which stays under NVIDIA's open model license. We'll say here exactly what happens to your fine-tune before our first Build is signed.

There's nothing to buy: the server stays ours until month 36, inside your monthly price. A monthly service fee is usually treated as an operating expense. Ask your CPA.

No. We run it. Remote access is outbound-only and logged, and your IT provider can watch every session.

Before your first desk beats its baseline, you can stop and owe nothing more. After that, leaving early means paying off the rest of the set-up share, not the full price. Either way you keep your data and a full export, and the software can go read-only at $750 a month.

It isn't built to cut jobs, and we never sell it that way. It's built to take the paperwork around the work off people's plates: the retyping, the forms, the handover notes. It never scores anyone. Your people spend more of the day on the work only they can do.

Its monitors check that it's actually answering, not just switched on. On your hardware, we can power-cycle it remotely, and your plan includes up to two on-site visits a year, anywhere in the lower 48 states. We'll publish uptime once we've measured it.

Not yet. We don't take controlled (CUI) or ITAR work until our defense package is ready. For context, CMMC Phase II was suspended on 13 July 2026, while self-assessments and NIST SP 800-171 still apply (Holland & Knight). Our reading; confirm with your counsel.

The founder for now, and customers once we have them. No outside investors. Customers pay one monthly price, in advance, from month 1.

No. We flag AI clauses in your customers' terms for your counsel and tell you how we read them.