Private AI
Private AI for small and mid-sized businesses
AI that runs on your own hardware, answers from your own documents, and never sends them anywhere. Built, installed and supported from Horsham, West Sussex.
Most AI a business can buy today is somebody else's computer. You type, the text travels, and what happens next is governed by a policy rather than by anything you can see. For a lot of work that is a perfectly reasonable trade. For the documents that carry your prices, your methods, your drawings and your customer history, plenty of owners would rather it did not leave the building — and now it does not have to.
On-premise AI, local LLMs, self-hosted AI — the same idea
These phrases all describe the same arrangement from slightly different angles. The model runs on a machine you own. Your files are indexed on that machine. Questions and answers stay inside your network. Nothing is sent away to be processed, and nothing you ask becomes training data for somebody else's product.
What has changed recently is not the idea — it is that open-weight models got good enough, and small enough, that a firm of twenty people can run one properly without a data centre.
What businesses actually use it for
- Asking questions across years of documents — quotes, drawings, certificates, contracts, job files. The answer comes back with a citation to the file it came from.
- Reading the awkward formats — PDFs, scans, spreadsheets, email attachments; the material that is technically filed and practically unfindable.
- Answering the question that would otherwise interrupt somebody — the shape that quietly decides whether know-how survives a retirement.
- Giving staff an AI they are allowed to use — which is the honest fix for people pasting company information into public tools. More on that.
Cortex
Cortex is our appliance: a GPU machine plus the software that indexes your shares and answers questions about them, with per-user permissions so people see only what they should. It is bought outright, from £4,950 installed. If you never pay us another penny it keeps working.
The demonstration is short and it is the same every time: we bring one, point it at some of your documents, and you ask it a question you already know the answer to. Then we unplug the internet and you ask another.
When renting is the better answer
Often, and we would rather say so here than on a call after you have spent money. If a few people use AI occasionally to tidy an email, and nothing sensitive goes into it, a per-seat subscription is the right purchase. We have worked the three-year comparison out honestly, including the cases where renting wins.
Owning starts to make sense when the AI needs to see the things you would not paste into a public box — and when enough people need it that per-seat fees have started to compound.
Where to start
With a measurement rather than a proposal. The Operations Audit puts a number on where the hours actually go and says plainly which of them AI would not fix. You keep the report whoever you hire.
Book a free 15-minute call Or send a message
Questions people actually ask
What is private AI?
AI that runs on hardware you control, using models installed on that hardware, so the documents you ask about never leave your network. The opposite arrangement is the usual one: you type into somebody else's website and your text travels to their servers.
Is private AI the same as on-premise AI?
Near enough. On-premise or on-prem means the machine is physically in your building. Private AI is the broader idea — it also covers a server you rent that nobody else shares. We do the on-premise version, because for a business of your size a box in the cupboard is cheaper and easier to reason about than a rented rack.
What is a local LLM?
A large language model — the technology behind ChatGPT — running on your own computer instead of somebody else's. The open-weight models available now are good enough for the work most businesses want doing: reading documents, answering questions about them, drafting and summarising.
Does it still work if the internet goes down?
Yes. That is the demonstration we do on site: unplug the network cable and ask it another question. It answers, because everything it needs is already in the building.
Is self-hosted AI automatically more secure?
No, and we will not pretend otherwise. A badly configured server in your building is a badly configured server. What owning it changes is that the boundary is yours to inspect and test, rather than described to you in a supplier's policy document.
What does private AI cost?
Cortex starts at £4,950 installed, bought outright, with optional Care from £195 a month. See pricing, or the three-year comparison against renting.