A fully configured machine with an open language model on it — plug it in, sign in, start working. No cloud, no US provider. Your clients’, patients’ and employees’ data never leaves your network. Pull the network cable and the system keeps working.
Professional secrecy outweighs any cloud provider’s assurance.
The EU AI Act’s high-risk duties were postponed in July 2026 to December 2027; the Article 50 transparency duties have applied since 2 August 2026. What that leaves untouched is what weighs heavier for law firms, practices and tax advisers anyway: professional secrecy and Article 28 GDPR. Hand such data to a cloud service and you are documenting someone else’s infrastructure — and you need a contract for it. With a machine in your own building there is no third party, and the AI Act file you keep later describes your own system.
/ 01
Professional secrecy stays intact
Client, patient and HR records are not handed to third parties. For law firms, practices and tax advisers that is not a preference — it is professional law.
/ 02
No processing abroad
No contract with a US provider, no third-country transfer debate, no renegotiation when a provider changes its policies.
/ 03
No vendor lock-in
Open models keep running even if a provider raises prices or retires a model. What sits on your machine belongs to you.
/ 04
Nobody reads along, nobody trains on it
With a cloud service your input passes through someone else’s systems: abuse detection that staff can look into, retention periods you do not set, and terms of use that can change. On your own machine those questions do not arise — there is nobody who could read along.
What's included
Ready to plug in means: you install nothing.
You obtain the machine to our specification — one you already own works just as well. We hand over a system your team can use the same day.
/ 01
Model pre-installed
An open language model, selected for your industry and hardware and tuned for the machine. European models on request.
/ 02
Browser interface
A chat interface reachable across your company network — like a familiar AI service, but on your machine. Nothing to install on workstations.
/ 03
User accounts and permissions
Per-employee access, encrypted connections on the network, separated areas where needed.
/ 04
Your documents as a knowledge base
On request we connect shared folders so the model answers from your own material — contracts, guidelines, manuals.
/ 05
Compliance dossier
System description, data-flow diagram, intended purpose and logging concept — the paperwork for your AI Act file, filled in.
/ 06
Team training
A session where your team learns what the system is good for and what it is not, with examples from your day-to-day work.
How it works
From order to a running system
01
A short conversation
We establish what you want to do with it, how many people will use it and whether hardware already exists. That determines the tier.
02
You buy, we set up
You order the machine to our specification — on request we place the order in your name. Then we set it up: model, interface, accounts, document access.
03
Install and train
Connected to your network on site or remotely, team trained, compliance dossier handed over.
04
Ongoing operation
On request we handle model updates, security updates and monitoring — otherwise the system is entirely yours from here.
Tiers and pricing
A fixed price — before the project, not after.
Starter has a fixed price. For Business and Enterprise we name a starting price, because the scope grows with your systems — we pin it down in the first call, and after that it does not move. You buy the hardware yourself: it is yours from day one, with warranty and guarantee directly from the dealer. We specify the requirements and, on request, order it in your name.
Would rather not deal with procurement? We order the right configuration in your name from the dealer. Invoice and guarantee run directly to you — all you have to do is take delivery.
A fixed price once the scope is set — independent of the hardware market
The hardware is yours from day one
Warranty and guarantee directly from the dealer, not routed through us
We also configure hardware you already own
The system and model are 100% yours
No mandatory subscription, care contract cancellable monthly
Why you buy it yourself: memory is scarce and expensive in 2026 and prices move sharply from quarter to quarter. If we resold the hardware we would have to price that risk in — and you would pay for it. This way you buy at the day’s price, get warranty and guarantee directly from the dealer, and our price stays stable. The hardware figures shown are guide values, not quotations from us. We recommend AMD because Apple currently offers nothing above 96 GB per machine — but we set up Apple and NVIDIA just as readily. On AMD there are two classes to choose from: a mini PC is the cheapest route to 128 GB, while the HP workstation costs around €1,800 more and brings three years of on-site warranty with it. If the machine is going into production use, that difference is worth knowing. The configurator below lets you compare the platforms directly.
All prices exclusive of VAT. This offer is directed at business customers only. Our Terms & Conditions apply.
Configurator
What do you actually need?
Set how large the model should be and how many people will work with it at the same time. You immediately see which machines are needed, roughly what the hardware costs and what our work costs on top.
Platform
Choose a machine
All figures are approximations to give you a first feel, not a quotation. What you actually need depends on the tasks you give the system and how long your documents are. We work that out in conversation.
Honestly
When PrivateLLM is not the right answer.
We researched and published on exactly this question — and for many companies we conclude that a cloud solution makes more sense. That holds even where it costs us a sale:
You are one or two people with occasional usage — a cloud subscription is cheaper and always current.
Your work needs absolute top-tier reasoning — open models are catching up but still trail on the hardest tasks.
Usage is irregular and hard to predict — then a purchase does not pay off.
Nobody on your side can keep an eye on a system and you do not want a care contract — then the box becomes the problem.
For everyday work — summarising, drafting, classifying, searching your own documents — a good open model gets very close. On the hardest reasoning tasks the large cloud models are still ahead. Before you order we tell you honestly which category your use cases fall into.
Which model gets installed?
We decide together based on your tasks, your language and the chosen hardware. Only open models you are permitted to use permanently — European ones by preference on request. Because the model landscape moves fast we do not commit in advance; we install the best fit available at delivery.
What happens when a better model appears?
With a care contract we swap it once the change is worth it — the hardware stays the same. Without one you can do it yourself; we hand over a guide.
Do we need a server room?
No. The machines are quiet and power-efficient and fit in a normal office cabinet. You need a socket, a network port and a spot where nobody unplugs it.
Does really nothing leave the building?
Not as delivered. The system works without an internet connection; we open it deliberately for updates. We ship purely local by design — a cloud connection only exists if you explicitly ask for it.
What if the machine fails?
Because you bought the hardware yourself, warranty and guarantee run directly through your dealer or Apple — not routed through us, which usually speeds things up. We assist with the process. Since configuration and knowledge base are backed up, a replacement machine is quickly back to the old state.
Can we test first?
Yes. On request we set up a time-limited test environment with a comparable model so you can check the quality against your real tasks before buying.
AI that stays in the building.
Tell us what you want to do with it and how many people will use it — we will tell you honestly whether PrivateLLM fits or whether a cloud solution serves you better.