Your own AI assistant, and your data never leaves the office.

Why we built this

Over the past year the question our clients have asked most has not been about networks or servers. It has been some version of: how do we use AI without handing our clients’ information to somebody else?

The honest answer used to be that you could not, not properly. That has changed. Open AI models are now good enough for real office work and can run on hardware a small business can own. So we built a private option: designed, built and supported by the same engineers who already look after offices’ networks, and sold as a complete system rather than a kit of parts.

Why businesses are asking for this

Law firms, accountants, medical and financial offices, and anyone with confidential client files have the same problem: the useful AI tools are cloud services, and pasting a client document into one means sending it to somebody else’s servers under somebody else’s terms.

Staff use them anyway, quietly, because they save hours. A private server gives them the same kind of tool with the data kept where it belongs — and gives you a written answer when a client or an insurer asks where their information goes.

Where your data actually goes

With a cloud AI service, everything a member of staff types or uploads is sent to that company’s servers and processed there. Free and personal accounts may use it to train future models unless somebody switches that off. Even business plans that promise not to train on it typically keep it for a period, hold it under terms the provider can change, and can be compelled to hand it over.

The larger risk is the use you do not see: staff pasting client details into a personal chatbot account because it saves them an hour. Once it is pasted, it is out of your control.

A private server removes the outside company altogether. Questions, documents and answers stay on hardware in your office. Nothing is used to train anyone else’s model, nothing sits on a third party’s servers to be breached or subpoenaed there, and you decide how long anything is kept — conversation history can be switched off entirely. Access is limited to your network or your VPN, with an individual login for each person.

For firms with confidentiality obligations

Law firms owe clients confidentiality, and bar guidance on generative AI expects lawyers to understand where client information goes when they use it. Medical and dental practices can only share patient information with outside services that sign a business associate agreement, which most consumer AI tools will not. Accountants and financial advisers handle tax returns, account numbers and identity documents every day.

Keeping the AI inside the office makes those obligations much simpler to meet, and gives you a straightforward answer when a client or an insurer asks where their information goes. It does not replace your own policies or professional advice on compliance — how the tool is used still matters — but it takes the biggest unknown out of the picture.

What you get

A dedicated machine, sized for your office, running current open AI models entirely on your own hardware. Your team uses it from a web browser, the way they would use ChatGPT: ask questions, draft and rewrite emails and letters, summarise long documents, pull key points out of contracts and reports.

It can also be pointed at your own documents — procedures, templates, past work — so staff can ask questions about them and get answers drawn from your files rather than the internet.

Accounts are per person, access is limited to your network or your VPN, and nothing is sent to an outside AI company. There is no per-user monthly subscription.

How the build works

It starts with a conversation about what your people would actually use it for, because that decides how powerful the machine needs to be. Then the hardware is specified, built, and installed in your office or comms closet.

Setup covers the AI models, the web interface, user accounts, secure access, backups of its settings, and a short session showing staff how to use it well. Everything is documented and handed over, and the server is yours outright.

Support after the build

Every build is handed over working, documented and yours. What happens after that is your choice, and you can change your mind later.

Look after it yourselves, and call when you need us. No monthly commitment. When you want something — a newer model installed, more staff added, a new set of documents loaded, a question answered — it is billed at $145/hr remote or $185/hr on-site. Suits offices with somebody technical in-house, or light use.

Keep us looking after it. The $650/month retainer puts the server under the same care as the rest of your network: monitoring so a problem is noticed before your staff notice it, security and software updates kept current, its settings and user list backed up, and the documentation kept up to date. You also get people who already know your setup when you need something, scheduled quickly. Hands-on work beyond that — model upgrades, new document collections, extra training — is billed at the published hourly rates or quoted up front.

Whichever you choose, the AI world moves fast. Better open models appear every few months, and part of looking after a private AI server is telling you when one is worth switching to — and when it is not.

If the budget is smaller

The $12,500 build is the one we recommend, because it is sized to stay useful as AI models improve over the next several years rather than just the current ones. It runs capable models quickly for a whole team, includes search across your own documents and per-person access controls, and has headroom for more staff and heavier use.

Lower-cost builds are available. They are genuinely useful, but the savings come from somewhere: smaller and less capable models, slower answers when several people use it at once, fewer of the features above, and less room to run the larger models that are coming. Some offices start there and upgrade later; others find it is all they need. Either way the differences are set out in writing before anything is ordered.

Honest limits

A private server will not match the very largest cloud AI models on the hardest questions. For everyday office work — drafting, summarising, searching your own documents — the gap is small, and for confidential material the trade is usually an easy one to make.

It is also a piece of equipment: it uses power, needs updates, and should be looked after like any other server. That can sit in the monthly retainer or be handled as needed. If your work involves nothing confidential, a paid cloud AI subscription may simply be cheaper, and you will be told that on the first call.