Private AI for UK firms
We supply and install a private AI system on a machine in your office, tuned to your documents — then keep it on the newest models every month. No cloud. No data sent to OpenAI, Google or anyone else.
FIG. 1 — Where your data goes
Switch between them to see where each question goes.
Data leaving your buildingIllustration
0 bytes
0 bytes
sent to outside AI
Your machine
runs on hardware you own
Monthly
upgraded as new models launch
Yours to keep
stop the retainer, keep the system
Section A1: Right now, most firms think there are two options. Both are bad.
Option 1
RejectedBan it
Your people fall behind firms that use it. And a ban is hard to police: under deadline, staff quietly paste a client file into a public chatbot on their own phone — and nobody sees it happen.
Option 2
RejectedAllow cloud AI
Every question, and every document pasted in, is sent to another company’s servers — often in another country, under that country’s laws. Client files, contracts, patient notes, candidate CVs: out of the building.
Section A2: The third option: AI that works where the files already are.
Never Leaves puts the AI on a machine inside your office. Your staff ask it to draft, summarise and search — the way they would use ChatGPT — from their own computers.
The model runs on that machine. Your documents and questions are processed there. Nothing is sent to an outside AI company.
| Cloud AI | Never Leaves | |
|---|---|---|
| Where the AI runs | Another company’s servers | A machine in your building |
| Where your questions go | Out over the internet | Down the corridor |
| Who processes your documents | The provider | Your own machine |
| Who owns the hardware | The provider | You |
| If you stop paying | Access ends | It keeps working |
Section A3: What you get.
Three parts. The first two are yours outright; the third keeps them current.
Part 1
The machine
A computer that lives in your office — a server cupboard, a comms room, under a desk. We spec it to your team size and the work you do. We supply it, or we install on suitable hardware you already own.
Yours. It stays in your building.
Part 2
The model, tuned to your documents
An open-weights AI model — the kind released by Meta, Mistral, Google, Alibaba and others — installed on your machine and connected to your own files, so it answers from your documents, not the internet’s.
- Drafts from your own precedents
- Summarises long documents
- Finds the paragraph you need
- Answers questions about your files
Part 3
The monthly upkeep
Better open models are released all the time. On the retainer we install them as they arrive, test them on your own work first, and switch over only when they do your job better. Security updates and support included.
- 01New open model released
- 02Tested on your own work
- 03Switched over — only if it does better
Section A4: Built for firms that hold other people’s information.
Typically 5 to 150 staff. A few of the things your team could ask it — on your own files, on your own machine.
Law firms
- Summarise a 200-page bundle into the points that matter.
- Find the termination clauses across a matter’s documents.
Accountants
- Draft the client letter from the working papers.
- Answer “where did we note that?” across a year of files.
Financial advisers
- Turn meeting notes into a first-draft report.
- Summarise a client’s file before the review meeting.
Clinics & private healthcare
- Draft the referral letter from the consultation notes.
- Summarise a patient’s history before the appointment.
Recruiters
- Compare a candidate’s CV against the job spec.
- Draft the candidate summary for the client.
HR & payroll bureaus
- Answer policy questions from your own handbook.
- Draft the letter from the case notes.
Engineering & manufacturing
- Search years of specs and project notes in plain English.
- Draft the method statement from past projects.
Section A5: How it works.
- 01
Consultation
A private conversation about your work, your files and what you want AI to help with. If it isn’t a fit, we say so.
- 02
Choose where it runs
On-site: we size a machine to your team and supply it, or use hardware you already own. Basic: we set up a private server that only your firm uses.
- 03
Install and tune
We set it up, connect it to your documents and tune it on your own work. Your staff use it from their own computers.
- 04
Every month: newer models
When a better open model is released, we install it, test it on your own work first, and switch over only when it does better.
Repeats every month on the retainer
Section A6: Clear prices. No surprises.
Basic · private server
£2,000–£3,000one-off
Your own AI on a private server that only your firm uses. Nothing to install in your office.
- Your data never reaches an AI company
- Never used to train anyone’s model
- Move to On-site whenever you’re ready
On-site · in your building
£5,000–£10,000one-off
A machine in your building. Nothing leaves it. Spec, installation and tuning on your documents; where you land in the range depends on:
- How many people will use it
- How powerful the machine needs to be
Plus the machinefrom about £3,000
If we supply it. The spec depends on your team and your work. Already have suitable hardware? You buy nothing extra.
Your figure comes out of the consultation.
Retainer
£300–£400per month
The same for both packages. Keeps it current and looked after:
- New models installed as they are released
- Each one tested on your own work before switch-over
- Security updates
- Support
If you stop
On-site: stop the retainer and you keep the machine and the model — they keep working. Only updates and support stop.
On Basic, everything that’s yours is handed over within 14 days — your documents, your settings and your AI set-up — and then our copy is deleted.
Section A7: Straight answers.
Does it need the internet?
On-site: running it doesn’t. The model works on the machine in your building, so asking it a question sends nothing out. New models do have to be brought onto the machine when we upgrade it — we agree with you how that is done. Basic: yes — your staff reach your private server over an encrypted connection. It still never goes to an AI company.
What’s the difference between Basic and On-site?
Where the AI runs. On-site puts a machine in your building, so your documents never leave it. Basic runs the same kind of private AI on a server that only your firm uses, so there’s nothing to install in your office — your documents leave the building to reach it, but never reach an AI company. The monthly retainer is the same, and you can move from Basic to On-site later.
Is it as capable as ChatGPT?
It depends on the task, and we won’t pretend otherwise. Today’s open-weights models are strong at document work — summarising, drafting, finding and answering from your files — which is most of what firms want AI for. We choose the model for your tasks and test it on your own work before your team relies on it.
Which AI model do you install?
Open-weights models — the kind released by Meta, Mistral, Google, Alibaba and others. We pick per task, and on the retainer we change it when a better one comes out.
What if we already have a server?
Tell us at the consultation. If it is up to the job, we install on it and you don’t buy a new machine. If it isn’t, we tell you what’s missing.
What happens if we cancel?
On-site: you keep the machine and the model, and they keep working. Only the monthly model updates and support stop. Basic: everything that’s yours is handed over to you within 14 days — your documents, your settings and your AI set-up — and then our copy is deleted.
Who can see our data?
Only the people you give access to. On-site, your documents stay on your machine in your building; on Basic, they stay on a server only your firm uses. Either way nothing is sent to an outside AI company. We only work on the system when you let us in for setup and support.
How long does setup take?
It depends on your documents and which package you choose. We agree a timetable at the consultation.
Section A8: Book a private consultation.
Four short steps, about a minute. We reply by email to arrange a time.