Skip to content
AI Tutor

How it works

Hints, not answers.

Studies of AI tutors found the same thing: tutors that held the answer back and guided learners step by step helped them learn. Chatbots that handed out answers did not, and in one large study learners who used one did worse in their exam.

  1. Step 1

    Safety first

    Every message is screened by a Nemotron safety check before the tutor sees it. If you write about being hurt or wanting to hurt yourself, you get a fixed message pointing you to a trusted adult and to Childline on 116, never a made-up reply.

  2. Step 2

    Marked against the mark scheme

    Each question has a hidden mark scheme. Nemotron checks your message against it, one point at a time, and notices common mistakes. Final answers in maths are checked by code, not by the AI.

  3. Step 3

    Hints first, answer last

    The program, not the AI, decides what happens next: ask you a question, give the next hint, or show one worked step. The full answer only opens once you have used every hint and step, and you cannot talk the tutor out of that rule.

  4. Step 4

    Checked before you see it

    Each reply is checked for the answer and for points you have not reached yet. If it gives too much away, it is rewritten, or replaced by a hint written in advance.

Made for phones

  • It runs in the phone's browser. There is no app to install and no account to create.
  • Pages are built in advance and the maths is drawn on our server, so the phone downloads less.
  • Large text, dark mode and reading aloud are in the settings at the top of every page.
  • A maths keypad gives you × ÷ ² √ π and more without hunting on the keyboard.
  • Prefer to talk? Tap “Talk to your tutor”. It reads the question to you, then you speak, it answers out loud and listens again. While it works out a reply it says so, so you are not left in silence.
  • Prefer paper? Take a photo of your working. Nemotron reads it into the box, and you fix anything it misread before you send.

Your privacy

  • You do not need an account. Your progress is saved in this browser only.
  • We do not store your messages. They are sent to the AI models to get a reply, with data retention switched off.
  • Reading aloud uses your device's own voice, so nothing extra is sent anywhere.
  • In talk mode the tutor's replies are spoken by a voice service (ElevenLabs or Murf). Only the tutor's words are sent to it, never yours.
  • In talk mode your voice is turned into text by your browser's speech service. We never record or keep audio.
  • A photo of your working is read once to turn it into text, then thrown away. We do not keep it.
  • The real-life examples after a question come from a web search for the question's topic. Your messages are not part of that search.

Built with

NVIDIA Nemotron open models, served by Nebius Token Factory: Nemotron 3 Ultra writes the replies, Nemotron 3.5 Super VL marks your work, checks each reply and reads photos, and Nemotron 3.5 Lightning screens every message for safety. When you finish a question, “Where you'll see this” shows where the idea turns up in real life, found with a live Tavily web search of trusted education sites. Built by Statotech Systems in Harare. The code is open source under the MIT licence.

Start a question