SUSI · Your own AI for business
Your own AI for businessYour knowledge. Your name. In Europe.
Anyone who wants their own AI almost always wants the same thing: it should know their own documents. It should carry their own name. And the data should not end up somewhere unknown.
See SUSIBased in Upper Austria. Servers in the EU. Language model from Mistral in Paris.
01 — THE SITUATION
AI has arrived in everyday work.
More than half of the people in work in the euro area use AI on the job. Two years ago it was a quarter.
Share of people in work who use AI on the job.
Around 20,000 respondents in eleven euro area countries, Austria among them. Source: European Central Bank, Consumer Expectations Survey, 26 August 2026.
Those who use it save a median of three hours a week — nearly eight per cent of their working time.
The situation in Austria
The same question, put to companies. The figures come from the business survey of the Information and Consulting division of the Austrian Federal Economic Chamber (WKÖ), carried out by KMU Forschung Austria. So those asked are members of that division — the IT and consulting sector, not the Austrian economy as a whole.
Business owners in the Information and Consulting division of the Austrian Federal Economic Chamber. Source: KMU Forschung Austria on behalf of the WKÖ, 27 April 2026.
Not even half — and that in the IT and consulting sector, where you would expect it most. In between lies the difference between used and introduced: many companies do not officially know that their people have long been working with AI.
So it is no longer about whether AI comes into the building. It is here. The only question left is whether it runs in an orderly way or on the side.
02 — THE RIGHT APPROACH
Three steps to AI in a company
Before it is about providers and models, it is about you. Structured and planned — this is where we can make the difference.
What should AI actually do?
Before you start, it has to be clear what is supposed to come out. We do not bake a cake by sticking candles into eggs either — the cake in between is missing.
Test it
A trial installation in your own operation, with the processes you actually intend to use. Not with an example from a presentation.
No lock-in
Keep the investment as low as you can. AI keeps developing — today one is ahead, tomorrow another. You have to be able to switch without starting over.
And then we come in.
We connect any model you want. Susi is developed further continuously. That way you commit neither to one model nor to one state of the art.
03 — THE TERM
What “your own AI” means in practice
The term means two different things, and everything else follows from which one.
The Susi model: An AI that knows your company knowledge, carries your name, is reachable at your address and keeps your data separate.
The alternative: Your own language model on your own hardware, trained on your own data. That is possible. What it looks like follows below.
What your own hardware means
- Models that can sensibly be run on hardware you can afford to operate are many times smaller than the large ones. The answers are accordingly.
- Buying it is the smaller part. Operation, electricity, updates and someone who knows the subject keep running.
- A trained model is a still image. If your product documents change, you have to retrain.
- At the next leap in models you start over again.
One advantage remains
With your own hardware the data stays in the building, under your own control. That is a real argument — it just applies to very few today. Anyone using Microsoft 365, SharePoint, OneDrive, Dropbox or any other cloud service has long since stopped keeping their data under their own roof. Exceptions do turn up, though not particularly often.
We do not build an AI. We build the framework around an existing one and give it your name, your look, your address and your separate data storage. That is a lot — but it is something else, and the difference belongs on the table before you decide.
04 — THE COMPARISON
Both options side by side
On the day it goes live, your own model looks good. After that it comes down to who keeps it up there.
| Your own AI hardware | With us | |
|---|---|---|
| The model | As big as your hardware can carry | The strongest the EU buildsOr an American one, if that is what you want. Susi is flexible. |
| Does the model get better | New hardware, set it up again | You are working with it automatically, shortly after release |
| Example: new product documents in the building | Retrain | Update the library with a few clicks, done |
| What comes up continuously | Operation, electricity, updates, someone who knows the subject | Usage |
| Switching to a different model | From scratch | We connect it |
For your staff to use your AI consistently, it has to be state of the art. A model that clearly lags behind current developments causes dissatisfaction and leads to other models being used past your own AI again. When a large model advertises new features, people listen — and many of them have it installed on their phone.
With us you pay for usage, not for upkeep.
05 — FREE AND PRIVATE
Free access and private accounts
What we often hear when we talk about AI in companies:
- “Our staff only use AI now and then, mostly with free accounts.”
- “Is there a free version?”
- “Our staff use their private accounts.”
- “We share one account in the department.”
Normal? Yes. Right? Well…
“Only now and then.” Anyone who rarely uses a tool never learns it. And anyone who cannot use it does not check the answer — they simply take it. Occasional use is not the harmless version. It is the unchecked one.
“Is there a free version?” Free is a plan, not a gift. It gets paid for anyway — with what goes in.
“With their private accounts.” Then whatever is created there belongs to the account and not to the business. Templates, histories, the tone that finally fits — all of it leaves when the person leaves.
“We share one account.” One account for several people means: nobody knows who asked what. The credentials are held by everyone who uses it — including those who will be working somewhere else next year. Whether a shared account is permitted at all is also in your provider’s terms.
06 — IN DAILY OPERATION
What this means in daily operation
If someone in your company uses a private or free AI account for work, you have AI. You just do not have it in hand.
The blog articles, the social media posts, even many emails and meeting documents. They are created with AI because it simply saves time. You should have control over that in any case, because:
- Nobody knows what was entered, or where it went.
- What someone has built up leaves with them when they leave.
- Five people, five accounts, five different answers to the same customer.
- It is paid for privately or not at all. The working time is yours either way.
“No AI in your company? Then ask your people what they wrote yesterday’s mailing and blog post with.”
Two questions. Which tool do your people use — and on which plan? What happens to your input is set out in the terms of the account actually being used. Those are the ones that apply, not the ones that would theoretically exist with the same model on a different plan.
07 — YOUR DOCUMENTS
How your knowledge gets in
Training teaches a model a tone, not knowledge. For knowledge it is the least precise route there is.
With Susi you set up libraries.
Your library is searched with every question, and the answer comes with a source reference — you can check what it is based on. A new document takes effect immediately. A removed one stops having effect immediately. Updating the library means: swap the file, done.
Your documents are held with us, in the EU. What goes to the model — when taking them in and with every question — is neither stored there nor used for training.
New product documents, Monday morning.
Your administration swaps one file in the company knowledge. From the next question onwards the AI answers from the new version — for everyone in the company, at the same time. Nobody has to pass anything on, nobody is accidentally still working from the old one.
The basis for it can be pulled from your website: on request it is read out automatically. Documents are added on top, as many as you like. The company knowledge is filled by your administration, and it is automatically available to every user.
Alongside that, everyone sets up whatever they need themselves. The colleague in sales has her own library of quotation templates and shares it with the two others in sales — she needs nobody’s help for that.
08 — YOUR NAME
Your name, your address
From the one-person business to the corporate group.
One project, everyone in it: the project lead assigns once, everyone picks up their own task, and the status is there for the asking.
An example: Fred’s Innovative GmbH gets an AI — Fredl.
Not “the AI tool”, not “the software from vendor XY”. Fredl. Reachable at fredl.freds-innovative-gmbh.at, in the company’s colours, with its logo at the top.
And Fredl works along. The colleague in purchasing has the French data sheet translated. The technician asks what the valve’s assembly instructions say instead of calling three people. The assistant turns twenty pages of meeting transcript into a one-page summary. And the new hire in week two asks Fredl — instead of not daring to ask.
When Fred tells his people to go through it with Fredl, everyone knows what is meant and where to go.
Hi Fredl — can you summarise the French data sheet for the V-200 in English for me?
Sure. The three points that matter for installation: operating pressure up to 16 bar, medium temperature up to 120 °C, connection G ½ inch. The rest is in the appendix.Source: data sheet V-200, page 2 — from your library
Thanks. And where do I find the current product documents?
In the company knowledge, as of 1 September. Shall I summarise what has changed compared with the previous version?
That creates identity. It is the difference between a tool that gets rolled out and a name everyone uses. That gets all your staff on board, and everyone knows where to find your AI.
Your own address and the shared company knowledge are part of White Label. We set that up on request: tell us what you have in mind and we will look at it together.
Ask about White Label09 — EUROPE
European, and what exactly that means
Based in Upper Austria. Servers in the EU. Language model from Mistral in Paris.
No Google services, no US cloud hosting, delivery via a European CDN, login protection on our own server. Your conversation histories are kept with us, not with the model provider.
Where Europe ends
When SUSI looks something up on the web, she asks Staan¹ — the European search index from Qwant and Ecosia, run in France under European law. Staan knows German, English and French.
What goes out is less than it sounds: the search runs through our server, with our IP address. What arrives at the search service is a search phrase the model has formed — no IP of yours, no browser or device details, not your conversation, not your documents and nobody who asked.
Where EU endpoints exist, we use them. They do not exist everywhere: image generation runs in the USA, with a provider we chose because it does not store your requests. If you do not want that, we switch it off for you.
¹ Staan has a fallback to the USA when there are no results on a topic.
What stays with the model provider
Nothing. What goes to the model — when taking in your documents and with every question — is neither stored there nor used for training.
You decide which model does the work.
By default Susi runs on Mistral from Paris. That is our decision for Europe, not a requirement we put on you. If you want a different model — an American one too — you get it, and immediately. In White Label as well.
What then applies to your data depends on that model’s provider. What we have agreed for Mistral applies to Mistral.
“These days I look things up in Europe, at Staan. Only for drawing do I still have to go over to America.”
Questions we get asked
What exactly does “your own AI” mean at your company?
Your documents, your name, your address, your separate data storage. No language model of your own on your own hardware — we do not build that, and further up you can read where the difference lies.
Can we work with our own documents?
Yes. You fill your library yourselves and keep it up to date yourselves. Every answer comes with a source reference.
Does it run under our own address?
Yes, with White Label — for example on your own subdomain, in your colours, with your logo.
What happens to our data at the model provider?
Nothing is stored there, nothing is used for training.
I want to use an American model — can I have that as White Label too?
Yes, and immediately. You tell us which model should work behind your interface. Nothing else changes: your name, your address, your company knowledge.
What then applies to your data depends on that model’s provider. What we have agreed for Mistral applies to Mistral.
How quickly are we up and running?
Getting in is a licence in the shop, getting out is one click at the end of the month. White Label runs through a request — we set that up together.
It is quicker and costs less than you think.
An AI that carries your company’s name.
From the one-person business to the corporate group.
See pricingLet’s talk