Built for schools
They're going to ask it for the answer.
It can't give it.
ThinkLM coaches students through their own reasoning on the problems you assign. The answer is sealed on the server before a student ever types — no trick, no jailbreak, no copy-paste can pull it out.
Third period
Two screens in the same room.
Mr. Rivera cuts a 19.6 m rope into 4 equal pieces. How long is each piece?
Answer · sealedMeanwhile, on your screen
Where the class stalled
- Choosing the operation 9 sessions
- Setting up the division 5 sessions
- Interpreting the result 3 sessions
How the class did — 18 sessions
Illustrative numbers — your readout is anonymous and class-level. School leaders see the same story school-wide: never a named student, never a per-teacher score.
She worked it out herself. You saw exactly where the class got stuck.
How it works
Type it once. It stays sealed.
-
Write the problem, type the answer
Create a problem in plain text — or upload a worksheet photo and let ThinkLM split it. You type the answer once, and a built-in check flags a likely slip before you approve.
-
ThinkLM seals it
The answer locks server-side. After sealing, even you can't read it back — that asymmetry is the anti-cheat guarantee.
-
Students join with a code
No accounts, no names, no downloads. A room code on your board is the whole rollout — one code can even carry a whole problem set.
The difference
Every AI tutor now promises it won't give answers. Ask where the promise lives.
The words are identical across the industry. The mechanism isn't.
The refusal
Elsewhere, a system prompt asks the model to hold back. Prompts can be argued with — industry tools measure answer-giveaway as a rate to reduce, not a door that's closed.
Here, the sealed answer never enters the conversation, and a guard checks every outbound sentence against it. There is nothing to argue with.
The problems
Elsewhere, a platform content library — or whatever the student pastes in.
Here, your own assignment, sealed with your verified answer. The tutor works your curriculum and checks a student's claim against your truth.
The students
Elsewhere, named accounts, rostering, and every chat stored for adult review.
Here, anonymous in class. No accounts to breach, no transcripts to subpoena — you get the class-level story instead.
Our own honesty line: the deterministic guarantee covers the sealed value. Everything else is guarded, judged, and fails closed — toward sealing more, never revealing more.
Your side of the wall
The refusal is student-side only.
Your side is a generative prep assistant that works for you — full lesson arcs, worksheets split into sealable problems, misconception coaching. The wall sits between teacher and student, exactly where cheating happens. These are real recordings of the app:
For school leaders
For the people accountable for the building.
Oversight without surveillance
A leader sign-in reads an anonymous, content-free readout across every class — engagement, stall points, mastery bands. Never a named student, never a per-teacher score.
Bands you already use
Results roll up as Masters / Meets / Approaches / Did Not Meet — the language your campus already speaks.
Accessibility on the record
WCAG 2.1 AA across the student and staff surfaces, with a VPAT available for procurement.
Policy-ready
Mapped against state K-12 AI policy frameworks, with the documentation to show for it — ask and we'll send the mapping.
Your data stays yours
Model providers are contractually barred from training on your school's content — and students have no accounts, so the riskiest data is never collected in the first place.
Nothing to install
Browser-only, on the devices you already have. Students join with a room code; staff sign in with a password your school controls.
A generation is learning to get answers without learning to think. Not in your school.
Built with real teachers, for one school first. It can hold yours.