Your office. Your AI.
Nothing leaves the building.

The confidentiality problem cloud AI can’t solve

Some information simply cannot be handed to an outside company, no matter how good its privacy policy reads. Attorney-client material, protected health information, unreleased financials, and proprietary client data all fall into this category for most regulated or client-facing businesses. Even “enterprise” tiers of the big cloud AI tools — the ones that promise not to train on your conversations — still route every message through a third party’s infrastructure, subject to that company’s outages, policy changes, breach history, and pricing decisions. A private AI server removes the third party from the equation entirely: there’s nothing to trust, because nothing leaves the building in the first place.

Ongoing cost, not just risk

The second driver is economics. Per-seat AI subscriptions add up quickly and never stop — ten employees at roughly $30 each per month is over $3,600 a year, forever, and the price is the vendor’s to raise whenever they choose. A local AI appliance is typically bought once, covers everyone in the office at no additional cost per login, and keeps running indefinitely, even without further payments.

Working without the internet

A private AI server keeps functioning if the office internet goes down, if a cloud vendor has an outage, or if a subscription policy changes overnight. The intelligence lives in the building, not in someone else’s uptime record.

The appliance is a compact, quiet computer that plugs into your existing office network like a printer or a NAS. It comes with the AI models already installed. When you turn it on, it starts serving a private web app at an address on your own network — something like youroffice.local — and anyone on the office Wi-Fi can open it in an ordinary browser and log in with their own account.

From there, every request — a chat message, a document upload, a request for a draft proposal — is handled entirely by the box’s own processor and memory. The AI model itself is a file that lives on the machine: it was trained on an enormous amount of text before it ever arrived at your office, and that learning is compressed into the model, the way a well-read colleague carries knowledge around without needing to look everything up. General knowledge and reasoning are built in from day one; anything current or company-specific comes from what your team uploads, or from the optional web search add-on.

You choose how much “thinking power” to run via the AI engine tier — a lighter engine for everyday writing and summaries, and a stronger one for harder reasoning and reading scanned documents or screenshots. Switching is instant on the appliance, because every engine is preloaded.

Two things are worth being explicit about, because they’re the whole point: first, nothing about a normal day of use requires the internet — chat, document answers, scheduling, and document generation all run entirely on the local network. Second, the only things that ever reach outside your building are things you deliberately turn on: the optional web-lookup add-on (and even then, only the search words leave, never your files), the optional remote-access tunnel, and a small licence/update check that never carries any of your content.

First-time setup, in order

The first screen you see asks for a handful of things: a licence key (software only — the box already has one built in), your company name to brand the workspace, an optional custom web address like yourcompany.privateofficeai.com so nobody has to remember an IP address, and finally the owner’s name, email and password. The AI starts on its lightest engine so it’s usable immediately, and you can move up to a stronger one any time from Settings.

Choosing an AI engine

Think of the engine as how much thinking power you’re dialing in. Swift is quick and light, great for everyday writing and knowledge-base answers. Balanced adds stronger reasoning and is the right default for most offices. Maximum is the deepest reasoning tier, and the only one that reads images, screenshots and scanned documents. A box has every engine preloaded, so switching is instant and fully offline.

Adding your team

From the Team page, add a person by name and email — they receive a temporary password by email and choose their own on first login. You can promote someone to Administrator, reset a password, or disable access instantly if someone leaves. Two roles exist from day one — Administrator and Team member — and you can add ready-made roles like Manager, Contributor, Basic user, or Guest, or build fully custom ones by ticking exactly the permissions you want.

Backups, updates, and what happens if the power goes out

The box backs itself up roughly once a day and keeps the fourteen most recent snapshots, stored on the box itself — nothing goes to the cloud, so a mistake stays recoverable without adding a privacy risk. If the power cuts out or the box restarts, it starts your office AI back up automatically the moment it boots. Every purchase includes a full year of free software and security updates, installed automatically in the background without interrupting a live conversation; after that, an optional Care Plan keeps them coming, though the box keeps working forever either way.

If something goes wrong

Most one-off issues clear with a simple restart of the box. Can’t reach it on the network? Confirm your device is on the same Wi-Fi and try its local name or network address from Settings. An add-on you just paid for not showing as active yet? Give it a few minutes, or check for it manually from the Add-ons page. Only the Maximum engine reads images — if a screenshot isn’t being read, that’s usually why. For anything else, real people are available by email, and Care Plan customers get priority support with remote fixes.