Could our IT provider just build this?

Standing it up, probably yes. The hard part is the four failure modes above, and none of them announce themselves. A filter applied after the search instead of before it works perfectly until the day it does not.

What happens if you are unavailable?

The deployment is in your account, on your hardware, defined in code you hold. It keeps running whether or not we do. What you would lose is the stewardship, not the system, which is worth pricing into the retainer.

How is this different from the legal AI tools we already see?

Those answer from the law, and they are better at that than we are. This answers from your files: the matters you already ran, the positions you already took. Different corpus, different question. Plenty of firms will want both.

Why not wait a year for it to get cheaper?

The model is the part that gets cheaper, and it is the part you swap. The boundary, the access rules and the audit trail are built once. If your documents are not in a state anybody could search today, waiting is the right answer and the assessment will tell you so.

Nucybersec: private AI that never leaves your account

Nucybersec builds retrieval AI that answers from your own documents and cites its sources, running inside a boundary you already control. Your own cloud tenancy, or hardware in your own building.

What it is

Two weeks. $4,500, credited against a deployment. You keep the findings report, target architecture, and cost model whether or not you proceed.

If you are comparing this against a per-seat assistant, we are not the right supplier.

Choose your next step.

The recorded briefing (Free - twelve slides, about five minutes, no sign-up). The trust briefing, narrated, with the full transcript on the page so you can read it instead of watching. Five more pages of reference material sit behind it: the capability map, the metric definitions, the requirements matrix, the data boundary and the sources.
The demonstration (60 minutes, live, in your building or online). The architecture the assessment scopes, running on one machine with the network unplugged. A real corpus indexed, searched by meaning rather than by filename, asked out loud and answering back. Nothing is installed on your side and nothing is asked of your IT.
The build session (Per engagement - half day, 4 to 12 people, on your own machines). The same stack, installed and running on hardware you own, by people senior enough to decide what happens next. You keep the compose file and the runbook, which is the recipe your IT would deploy rather than a souvenir.
Readiness Assessment ($4,500 - fixed fee · credited against the deployment). Two weeks. We audit your account and data estate against the controls this architecture enforces, and hand back the specific gaps between where you are and a defensible private deployment.
Sealed Deployment ($35,000 - fixed fee · 4–6 weeks · in your cloud tenancy). The framework deployed into your account, your network, your corpus. Assistant live, guardrails tuned to your content policy, observability wired to the alerting you already run.
Retrieval & Integration ($45k–90k - scoped · 3–4 months). Corpus at scale, your identity provider in front of it, your systems of record feeding it. Scoped rather than fixed, because the corpus decides the work.
Stewardship ($3,200 - per month · cancel anytime · no refunds). We hold the pager. Model migrations, guardrail tuning as your policy moves, quarterly posture review, and the cost report you cannot produce yourself.

What this does not do.

It is not multi-tenant, and never will be.
Neither deployment is a product you install yourself.
It is aligned to control frameworks, not certified against them.
Quantised indices leak membership signal, and we have not solved that.

Four questions, and the answers you should get.

Can it reach the internet at all?
Where do our documents go when someone asks a question?
Who can read what we ask it?
How would we know if that ever stopped being true?

The questions you are already asking.

Could our IT provider just build this?: Standing it up, probably yes. The hard part is the four failure modes above, and none of them announce themselves. A filter applied after the search instead of before it works perfectly until the day it does not.
What happens if you are unavailable?: The deployment is in your account, on your hardware, defined in code you hold. It keeps running whether or not we do. What you would lose is the stewardship, not the system, which is worth pricing into the retainer.
How is this different from the legal AI tools we already see?: Those answer from the law, and they are better at that than we are. This answers from your files: the matters you already ran, the positions you already took. Different corpus, different question. Plenty of firms will want both.
Why not wait a year for it to get cheaper?: The model is the part that gets cheaper, and it is the part you swap. The boundary, the access rules and the audit trail are built once. If your documents are not in a state anybody could search today, waiting is the right answer and the assessment will tell you so.

Nucybersec LTD is a North Carolina limited liability company. Software licensed per deployment; possession of a copy does not grant a licence to run it.