A trainer that counts only what you can do on your own.
Four AI agents build you a course, set you real work, check it without taking the Trainer’s word for it, and change the plan on the evidence. On an iPhone or a Mac it runs with no internet, no key and nothing to pay.
Help arrives in steps, and the steps are part of your record.
Every challenge has a hint ladder. Each rung you take is counted, so “mastered” means something: you did it, and we know how much help you had.
Independent
Done with no help. The only thing that counts fully toward mastery.
With minor hints
Counts, less. Shows the Trainer where to go next.
Guided or shown
Recorded so the plan can return to it. Never mistaken for mastery.
Four agents, one loop. You can read every word of it.
Curriculum Architectbuilds the plan
Turns your goal and your self-rating into a roadmap of skills, each one a thing you can do with a measurable standard. The whole plan is visible before you start.
“A skill is one thing a learner can do. Cover less of the subject rather than widen the skills to fit.”
Trainersets the work
Writes the challenge, answers your questions, hands out hints one rung at a time, and teaches first when you have said you are starting from nothing.
It shows the command and its output. One idea per challenge. Criteria state a standard, never the answer.
Assessorchecks independently
Grades the work without ever having set it. Where a program, a test suite or a checker can decide, it decides and no model is asked at all.
The setter is never the grader. That is the rule, not a preference.
Supervisorreviews the training
Looks at how the training itself is going: what stuck a week later, where the hints piled up, what the Trainer should do and stop doing next. Its findings bind the Trainer.
“Numbers held. Move to prices. Stop re-explaining greetings; they are solid.”
Made for you. Then remade as you go.
You decide what you want to learn and how you like to learn it. The four agents watch your training closely, adapt to your way of learning, and rewrite the plan when they see you are not comfortable.
You set the goal
Say what you want to be able to do, and how much you already know. Your course starts from there, not from chapter one.
You choose how you train
Writing, a code sandbox, a terminal, a conversation. Pick what suits you. The agents stay inside your choice.
The agents watch closely
Every answer you give, every hint you take and every question you ask is seen. They notice when you are stuck and when you are flying.
The plan changes for you
When something is hard, the plan slows down and narrows. When something is easy, it skips ahead. Every change is written down with its reason, so you can read why.
On your iPhone or Mac, it runs with no internet and nothing to pay.
The four agents run on Apple’s on-device foundation model, on the phone in your pocket or the laptop on your desk. Your answers stay there. There is no key to buy and nothing to subscribe to.
For a harder subject, switch a goal to Apple Private Cloud Compute, or to your own Claude, OpenAI or OpenRouter key. Your choice, per goal.
- No internet
- Train on a plane, on the train, in a village with one bar of signal. It works the same.
- No key
- Nothing to sign up for and nothing to paste in. Open the app and start.
- Nothing to pay
- The model is part of your iPhone or Mac. Running it costs you nothing.
- Private
- Your answers are read on the device and stay on the device.
Everywhere you are.
One account. Start on the iPhone, continue on the Mac. Goals, courses and progress follow you. Share a course, or import one someone shared.
iPhone
Speak your answers; transcribed and graded on the phone.
Mac
A real window, with room to write code and run it.
Windows
Import a course file, train offline, export your progress.
Web
In the browser. Nothing to install.