PRIVATE AI · FOR BUSINESS

Private AI for business. Your documents stay yours.

A private AI assistant for your company runs on servers you control: ours in the EU, or placed at your site. It uses open-weight models, so your prompts and documents are not sent to a public AI provider. We size it, set it up and run it. Price it below.

  • People
  • Sign-in
  • Model
  • Documents
  • Your perimeter
  • Public AI provider
  • Nothing is sent
Where the processing happens in the hosted and on-premise options.

THE NUMBERS BEHIND IT

Four things you can check.

SERVERS
1–8

A standard setup runs on one to eight servers, from a small team to a large company. Beyond eight, our team scopes it with you.

MEMORY PER SERVER
128 GB

Each server holds 128 GB of memory, shared by the model and every active person's working memory. A large model spans two to four servers; more people add servers.

PLACES TO RUN IT
3

Our EU servers, servers placed at your site, or a closed network. The estimate below prices all three.

PRICE SOURCE
Our price list

The monthly figures are our current prices. One-off work is a range we confirm in the formal quote.

WHY PEOPLE ASK

Three questions that come before the project.

None of them needs a technical answer first. They need a number or a decision.

Where do the prompts go?
A public AI assistant processes what you type on its provider's servers. For contracts, client files and internal minutes, that is the first thing to settle.
Which machine, for which model?
Nobody wants to order hardware before knowing what will run on it. The estimate below starts from the model and works down to the machine.
Who keeps it running?
Choosing a model, serving it, patching the server and updating the model later is real work. We do it, or we hand you the sizing if you run it yourselves.

WHERE IT RUNS

Three places to run it.

The same models in each. What changes is who owns the machine and who can reach it.

Hosted in the EU

We run the model on dedicated AI servers in the EU and look after them. One server or several, as many as your load needs. No hardware to buy.

Fits teams that want the result without owning the machine.

From €448 a month

At your site

The model runs on servers at your site, inside your network. We can supply and install the servers, or size them for you to buy. We install the stack and support it.

Fits companies that want the servers under their own roof.

From €49 a month in support, hardware not included

Closed network

For networks with no outside connection. The servers come from us or from you. This is scoped project by project, so the figure here is a starting point.

Fits environments where nothing may leave the building.

From €49 a month in support, hardware not included

YOUR DOCUMENTS

Answers from your own archive.

Search over company documents is a use case you can switch on in the estimate. It is built per project, because no two archives are alike.

We index what you choose
Contracts, procedures, minutes, tickets: you pick the folders, we load and index them on the machine that runs the model.
Access is part of the scope
We agree with you who may query which documents, and connect your sign-in if you add it in the estimate.
The index stays on your side of the line
In the hosted option it lives on our EU servers. In the on-premise and closed-network options it lives on yours.

RULES AND RECORDS

What private processing changes for GDPR and the AI Act.

Running the model on infrastructure you control makes some questions easier to answer. It does not replace your own assessment.

You can say where the data is
Processing happens on servers we run in the EU or on your own, so the answer to where is a place you can name.
You can say who can reach it
No public AI provider receives the prompts. Access is limited to your people and the team that runs the machine.
You still own the assessment
Risk classification and the records your sector requires stay with you. We help you gather the technical facts; we do not certify your use of AI.

SIDE BY SIDE

Private AI and a public AI assistant.

A public assistant is quick to start. The differences show up when the content is sensitive.

Differences between a public AI assistant and private AI
TopicPublic AI assistantPrivate AI with us
Where prompts are processedOn the provider's serversOn infrastructure you choose
Who receives your promptsThe provider, under its termsOnly you and the team that runs it
Works without a connectionNoIn the closed-network option
The modelThe provider's, changed when it decidesOpen-weight, a version you pick and keep
How the cost growsWith people or usageWith machines, once you outgrow one
Who runs itThe providerWe do, or you do, by the plan you choose

PRICE IT

Size it and price it.

Answer a few questions. The estimate updates as you go: a monthly figure for the infrastructure and support, and a range for the one-off work. Nothing is sent until you ask for the formal quote.

What will it be used for?

Pick everything that applies.

How many people will use it?
25 people
How many are active at the same moment?

Memory is sized on the people active at once, not on the headcount.

How much should it keep in mind at once?

The context window, in tokens: a token is about three quarters of a word. Longer context needs much more memory for each person.

Where should it run?
How capable should the model be?

Larger models reason better, need more servers and serve fewer people per server.

Power is how well it handles hard questions. Weight is how much of a server it takes up.

What should it connect to?

Optional.

How much support?

Get the formal quote

Leave your details and we send a written quote based on these answers. A person replies.

Your details are used to reply to this request. Read the privacy policy.

HOW IT STARTS

From the use case to a running assistant.

  1. 01

    Scope

    We list the use cases and the documents, and agree where it runs.

  2. 02

    Size

    We pick the model and size the machine with the method behind the estimate.

  3. 03

    Set up

    We install the stack and connect what you chose: sign-in, ERP, mail.

  4. 04

    Run

    We monitor it and update models when it makes sense, at the support level you chose.

QUESTIONS

What people ask before they start.

Does private AI work without internet?

In the closed-network option, yes: the model and the document search run inside your network, and anything that needs the internet stays off unless you add it. That option is scoped project by project.

Does our data leave the company?

Not to a public AI provider. On your own servers it stays on your servers. In the hosted option it stays on servers we run in the EU.

How much does private AI cost?

It depends on the model, how many people are active at once and where it runs. The estimate above turns those into a monthly figure and a range for the one-off work; the formal quote confirms both.

How is it different from a public AI assistant?

A public assistant processes your prompts on its provider's servers, with the provider's model. Here the model is open-weight and runs where you decide, so the prompts stay inside that setup. The comparison above lists the differences.

What about GDPR and the AI Act?

Processing on infrastructure you control, in the EU, makes questions about where data is and who can reach it easier to answer. The assessment of your use of AI stays with you; we do not certify it.

WHAT THIS PAGE DOES NOT PROMISE

What we do not promise here.

The estimate is a starting point. These are the things it does not settle.

A service level

The estimate carries no uptime or SLA figure. Support terms come with the formal quote.

A delivery date

We do not give a go-live time before we know your documents and your network.

A ready-made closed-network package

The closed-network option is scoped project by project, so its figures are a starting point.

Reserved capacity

The estimate does not hold a server for you. Availability is confirmed in the formal quote.

A certification

We help you answer GDPR and AI Act questions about the technical setup; we do not certify your use of AI.

Tell us what it has to do.

Price it above, or write to us with the use case. A person answers.