A plain, honest guide to what a child sees on screen — and how the same engine teaches any subject, in any language. Written for the grown-up in the room, and for schools deciding whether to use it.
This is a pilot. Every screen, rule and worked answer below is what the engine actually does — nothing here is a mock-up. A first set of Klasse-4 lessons is live today; more of the curriculum, and more languages, are on the way.
Every topic opens with a switch at the top: Classroom or At home. It changes how much is locked and who sets the pace — the lesson content itself is identical either way.
Built for a teacher projecting the lesson to the whole class.
Nothing is locked. The lesson's worked examples are all open, so the class can move at the teacher's pace and look at any step together.
The lesson stays in one language — the one the teacher is teaching in — with read-aloud narration available for the room.
Built for a child working alone, at their own pace.
The child picks their own reading language. Sections unlock in order: a short quick-check at the end of each section has to be right before the next one opens.
Get a quick-check wrong once and the answer is held back — the child is pointed back to the section to re-read it. Get it wrong a second time and the full worked answer appears, so nobody stays stuck.
Each topic is taught in one main language — German for the Hessen lessons — and that stays the core. Beneath it, every paragraph and every question carries an equivalent in the child's language — English today, more languages added the same way — written ahead of time by a person, not translated on the fly.
A child picks their own language from a menu. When it differs from the teaching language, that equivalent is there to plug in under each paragraph — one quiet tap at a time, so the child opens only the sentence that blocks them and the language they're learning stays front and centre. Choose the teaching language and nothing extra shows.
Read-aloud follows the same choice: it reads the lesson in the child's language, at a pace you can slow down. A child who can't yet read German can still hear and follow the maths — or the German lesson itself.
It's plug-and-play: a new language is just one more equivalent added to each block. The million-number maths chapter already carries an English equivalent under every paragraph, and the compound-interest lesson runs as a full parallel in English and German; more languages drop in the same way.
And before the lesson even begins, a short list of the chapter's key words appears — each German term with its meaning in the child's language — so a child meets the vocabulary first and isn't stopped later by a single unfamiliar word.
Separately from the language choice, a Simple ↔ Full reading-level toggle (the Lesestufe) switches between short, plain wording and the fuller text with its extra worked examples — the same lesson meets a struggling reader and a confident one where they are. It opens in Simple.
After the lesson, every topic has a “Your turn” section. Two kinds of question live here; the third — the test — makes up the final test further down. Both kinds give a child two tries, and the rule is the same: coach first, reveal after.
Number and multiple-choice questions mark themselves the instant a child answers — right in the browser, no waiting. A written answer (“write the sentence correctly”) is marked by the AI instead, and can be typed or photographed. First try wrong: the child sees what to do and which section to re-read — not the answer. Second try: the full worked answer appears.
Always marked by the AI, with the same two tries. The first try names exactly what went wrong and points to where it's explained, but holds the answer back. The second try reveals the worked answer. The feedback is specific — never just “wrong.”
The test questions cover every section of the lesson and make up the final test described below. They're written ahead of time — drafted with AI, then reviewed by a person — and served in rounds, never generated on the spot.
Because the marking is grounded in the lesson, feedback speaks the child's language literally and figuratively: it names the actual step that slipped and links straight to the section that teaches it, so the next try is an informed one — not a guess.
At the end of a topic sits the final test. It is a real check, but a formative one: it teaches while it marks. The child answers a round of questions and submits once; the test then works section by section, not question by question.
You cleared enough sections. Every worked answer is there to read through, even for the questions you got right — the test doubles as a last recap.
New questions come from just the sections still to clear, with a plain list of exactly which sections to look at again and a one-tap link to each.
Four examples of the feedback a child sees — leading with the universal one (compound-interest maths, understood anywhere), then German nouns — each with a right answer and a wrong one. In the final test, both the tick and the worked answer appear on every question, every round.
€1,000 grows at 10% a year. What is it worth after one year?
Child wrote: 1,100
Exactly. 10% of 1,000 is 100, so after one year you have 1,100.
📖 Section 2: One year's growth →€1,000 grows at 10% a year. What is it worth after one year?
Child wrote: 1,010
You added 10 — but 10% of 1,000 is not 10, it's 100. Ten percent means ten out of every hundred.
📖 Section 2: One year's growth →Which of these is a noun? Katze · grün · spielen
Child chose: Katze
„die Katze“ fits with an article, so Katze is a noun.
📖 Section 1: What is a noun? →Write it correctly: der hund spielt im garten.
Child wrote: Der hund spielt im Garten.
One naming word is still lower-case. Check which words are nouns — der / die / das fits in front of them.
📖 Section 3: Nouns are capitalised →Every answer gets feedback in simple language — one level easier than the lesson itself — so a child never has to fight the explanation on top of the subject.
At home, the same feedback is written a second time in the child's own mother tongue, chosen from a menu. The AI writes that translation on the fly, so a child who can't yet follow the lesson language can still read what to do next.
One thing stays fixed: the AI only puts into words what went wrong. The worked steps themselves — the actual maths, or the actual spelling rule — are written ahead of time (drafted with AI, then reviewed by a person) and shown verbatim, never invented in the moment. The marking model is a fast, cost-effective one, so a class of thirty stays affordable, and the pre-written answers are what keep it accurate.
A child never has to type. On any written question — a guided answer, a test answer, or a “write the sentence” practice item — they can photograph their handwriting (or attach a PDF), and a vision model reads and marks it exactly as it would mark typed text.
Where a child naturally works on paper — German writing, forming sentences, fixing capitalisation — the photo upload is front and centre: a drag-and-drop / camera panel, not a hidden button. Where the answer is just a number, it stays a quiet option. Photos are read to mark the work and are not stored.
No sign-up, no accounts, no names. A child never logs in, and we ask for nothing that identifies them. The only thing kept about a child is their own progress — which sections they have passed — and that lives in the browser on their device, nowhere else. Because there is no account and no name, we hold no personal profile of a child under the GDPR (in Germany, the DSGVO).
When a child submits an answer to be marked, the answer — or a photo of their handwriting — is sent to the AI marking model to be read and graded, and used only to write the feedback. We do not keep the photo or the answer afterwards.
The marking is done by Anthropic's Claude model. A school running a pilot can ask us for the full processing details and a data-processing agreement — the GDPR Article 28 processor contract, in German an Auftragsverarbeitungsvertrag (AVV).
We know the EU AI Act treats tools used in education seriously, with its high-risk obligations phasing in over the next years, and we are building toward them as the product grows.
Nothing in the engine is tied to maths, or to German, or to one country. A topic is just content in a file — its sections, its questions, its worked answers. The very same screens on this page teach all of these:
€1,000 growing at 10% a year, watched grow 1,000 → 1,100 → 1,210 → 1,331 — the same idea whether a child is in Hong Kong or Hessen, needing no cultural or language background to follow. Runs as a full parallel in English and German; the euro amount and the number style follow the chosen language — €1,000 in English, 1.000 € in German.
Place value, reading and writing big numbers, rounding — taught in German, every number stored once as a locale-agnostic token and formatted to the reading locale automatically (1.000.000). Tap any paragraph and the English equivalent sits beneath it — the same number written the English way (1,000,000).
Which words are naming words (der Hund, die Freude) and why German capitalises them. Written answers are typed or photographed and marked exactly like a maths answer — same tiers, same two tries, same section links.
To offer a topic in a new class language, the words are translated and checked. The structure, the marking, the section links and the numbers all come along for free.
The best way to understand any of this is to open a lesson in classroom mode, everything unlocked: read numbers up to a million the German way, or teach a child which words are nouns.