Smaller than a shoebox
The device sits on your own network. The language model lives on it, and so do the documents it works with. You decide which colleagues use it.
for notaries, accountants and lawyers
A language model that runs on a device in your own office. It reads your files, summarises them and writes a first draft, without internet and without a cloud service. You try it at our place first, with one task from your firm.
No obligation · 30 minutes · online, with someone who builds, tests and fine-tunes it
To be honest: we are still exploring local AI. We test what it can and cannot do today on our own machines in Leuven first.
At a glance
Sound familiar?
Cloud AI can work, with the right edition, contracts and settings. But with most providers a residual risk remains. For your most sensitive files, you would rather not have to weigh it.
ChatGPT at the office
Colleagues already use ChatGPT, “just for ordinary emails”.
An email about a case is never ordinary. One pasted paragraph now sits on a server you don’t know.
Your vendor’s AI
Your software vendor offers AI in the cloud, with a data processing agreement.
Even with data in Europe, American providers fall under American law. That is a residual risk you have to weigh yourself.
A ban
You simply forbid it.
Then the work stays where AI really helps: reading long documents, summarising, writing a first draft.
How it works
Local AI is also called private or on-premise AI: the language model runs at your place, not at a cloud service.
The device sits on your own network. The language model lives on it, and so do the documents it works with. You decide which colleagues use it.
You work in a chat window in your browser, just like ChatGPT. Only the answer comes from the device, not from a service on the internet.
No subscription per user. We install it, keep the model up to date and fine-tune it, if you want us to.
Who runs it
Universities and large companies run AI on their own servers, with their own team to manage it. A ten-person firm doesn’t have that department. We take that part on.
Computing power
Their own data centre with GPU servers.
One device on your own network, on one power socket.
Models
An in-house AI team chooses, tests and replaces the models.
We test a new model first. The model on your device doesn’t change without you knowing.
Access
An IT department manages who may use what.
We set it up in your firm’s name. Code, prompts and access belong to you.
Questions
An internal helpdesk.
You talk to the person who set up your device and your task.
Honest about the limits
Local models have become strong, but they are not the smartest models around. This is what we see in our own tests.
Does well
Not yet, or less well
Always
What AI writes is a first draft. Anything that leaves your office goes past someone who knows the file first. We record that per task in a one-page AI sheet.
Local or cloud?
| Local fits when… | Cloud fits better when… | |
|---|---|---|
| Data | You work with case files or personal data. | There’s no sensitive data involved. |
| The model | Tomorrow’s answer has to match today’s. | You need the smartest or newest model. |
| Your team | Many colleagues use it daily, with no price per user. | You’re a small team and mainly want to try things quickly. |
In your office
Also for GPs who want a consultation report transcribed, helplines where a conversation must not leave the building, and spin-offs whose unpublished research is the company.
Deeds and estates
Preparing a deed from your own templates. Searching an estate file for what you need.
Documents and questions
Reading in client documents and invoices. Searching your own internal notes when a client asks something.
Bundles and chronologies
Summarising a bundle of documents. Drawing up a chronology. Comparing clauses side by side.
How to start
Step 1 · 30 minutes
Together we pick one task and look at which documents it needs.
No obligation, online.
Step 2 · Trial
We set up the task with test files or anonymised documents. You see how good the answers are and how much time it saves you.
Real case files only go onto a device in your office.
Step 3
Your own device in the office, which we install and maintain. Or adjust. Or stop, if it doesn’t deliver enough.
You decide, with the numbers from the trial.
After the first conversation you get a proposal with a clear scope and a fixed price. If you choose your own device after the trial, we install it at your office and look after it: model updates, fine-tuning and support.
Who builds it
On our desk in Leuven sit an ASUS Ascent GX10, from the NVIDIA DGX Spark family, and Macs with Apple silicon. In 2008 a computer reached a petaflop for the first time: IBM Roadrunner filled a room and cost around 100 million dollars. Our device reaches the same number on one wall socket. It is not quite the same maths (Roadrunner calculated for science, this device for AI), but the leap is real.
We write down what works and what doesn’t. The dead ends too.
15 years
building and maintaining software
400+
organisations
4 years
using AI daily in our own work
Frequently asked questions
You can get a data processing agreement and a security overview on request.
Question not listed? Just ask us.
Not for everything. For searching, summarising and writing a first draft from your own documents it comes close. For complex reasoning without a source, the big cloud models are stronger.
Sometimes, yes. Business editions don’t train on your data and can process it in the EU, with a data processing agreement. But American providers fall under American law, even when your data is in Europe. For ordinary tasks that is often acceptable. For your most sensitive files, or as a lawyer following the guidelines of the Flemish Bar (OVB), a device in your office is the clearest choice. We help you choose per task.
Local processing removes the biggest risk, but you remain responsible for how you use AI. For each task you get a one-page AI sheet that you can show your professional body, DPO or insurer. This is not legal advice.
Yes. Summarising and writing in Dutch works well. Knowledge of Belgian law comes from your documents, not from the model itself.
You buy it once; the price depends on how much memory you need. After the first conversation you get a proposal with a fixed price for the device, installation and maintenance.
Local isn’t automatically cheaper than cloud AI. With a small team and little sensitive work, a subscription is sometimes the better deal, and then we’ll tell you.
We do, if you want: model updates, follow-up and fine-tuning. Code and access are in your firm’s name.
Tell us which task takes the most time. We will tell you honestly whether local AI can handle it today.
No obligation · 30 minutes · online, with someone who builds, tests and fine-tunes it