Companion
Tools that name their source.
Four applications on one foundation: fact answers questions with evidence, atrium takes in what arrives from your systems, learn turns your own material into finished teaching units, school puts the school day in order. Processed in the EU, released by a human.
The applications
Two worlds, four tools.
fact and atrium work for companies, consultancies and universities; learn and school for schools. The choice does not start with the tool — it starts with where you work.
Wirtschaft · Hochschule
Für Unternehmen, Beratung und Hochschulen.
Wissensarbeit mit Beleg und ein Hof, in dem die Fachsysteme andocken. fact beantwortet, atrium ordnet.
Schule · Unterricht
Für Schulen und Lehrkräfte.
Material, das zwischen zwei Stunden entsteht, und ein Schultag, der geordnet bleibt. learn erzeugt, school führt.
Why bother
A language model that will not say where it got something is an opinion.
What separates a tool from a toy is not the quality of the answer. It is whether the answer can be checked.
Every statement leads back
Answers carry source markers. One click opens the passage the sentence came from — not the document, the passage.
Your data stays where it belongs
European data centres, European models for transcription, nothing handed to training runs. What we do not need, we do not store.
A human signs off
Where people are involved, review is mandatory rather than optional. Whoever approves can see what the result rests on.
Choosing
Which one do you mean?
Deliberately coarse. A table listing every single feature does not answer the question that comes before it.
| fact | atrium | learn | school | |
|---|---|---|---|---|
| The question behind it | What do we know about this — and from where? | What has come in, and what needs doing? | What am I teaching with tomorrow? | How is this school day going? |
| Works on | Your documents, knowledge collections, web search | Mailbox, filing, business systems, receipts | Curriculum, open sources, your own material | Classes, timetable, marks, absences |
| Produces | A sourced answer, a report, a presentation | A structured record, a task, a note | A teaching unit, quiz, exercises, video | An entry, a mark, a message, an assignment |
| Companion app | none — runs in the browser | atrium for iOS | none — runs in the browser | school for iOS — the app is the product |
| Human release | Mandatory where people are involved | Before anything is written back | Before the class sees it | Marks and absences are set by the teacher |
| Availability | In use | In use | In use | In development — launching autumn 2026 |
What it looks like
Not a drawing. A screenshot.
Product pages like to show what would be possible. This shows what is there — captured from the running application, with the numbers the system held at the time.
On the move
Two apps for iOS.
atrium and school have a native companion. Not a second product and not a second set of data — the same records, the same rules, on the device that is already in your pocket. fact you use in the browser.
Common questions
What to settle first.
Where is our data processed?
In data centres inside the EU. Language models run through European providers or the European regions of the large ones; for transcription we use a European model by default. You get the list of services and processing locations before signing, not after.
Is our content used for training?
No. The models we use run without a training licence. That is in the data processing agreement, not just on a website.
Do we need our own IT department?
No. The usual route is hosted operation. For organisations with their own infrastructure there is an on-premise installation. We set up either one.
What happens when the AI gets something wrong?
It will. That is exactly why the chain of evidence is not decoration: every statement can be checked against its source in seconds, and nothing involving personal data leaves the system without human review. The question is not whether a model errs, but whether anyone notices.
Can we try it first?
Yes. We set up a test instance with your own documents rather than a demo dataset. A tool that only convinces on someone else’s examples has not convinced anyone.