A calisthenics app built by one person

Why it exists

The fitness apps around when this started fell into two groups. Some were catalogues of PDF routines with an interface on top: the same plan for everyone, with no way to tell whether it was too much or too little for you. The others were built for weights, where progress is easy to model because you only have to add kilos.

Bodyweight does not work like that. You cannot add two and a half kilos to a pull-up: you have to change the exercise, and picking the wrong next rung means stalling for months or getting hurt. That turns progression into a decision problem rather than an arithmetic one — and that is exactly the kind of problem a model can help with if you give it real data.

What was decided, and why

The 0–100 score uses an asymptotic curve instead of a linear scale. That is deliberate: in real life, going from 20 to 30 points of pushing is a few weeks, and going from 80 to 90 can be two years. A linear scale would have lied about that and made people feel stuck while they were actually progressing well.

The four chains are measured separately because a single "level" hides exactly what needs fixing. Almost everyone arrives with a large gap between push and pull, and not seeing it is why that gap is still there a year later.

The catalogue treats floor, wall and chair as universal equipment because the real barrier to entry is not motivation, it is not owning a bar. If the first routine you get asks for something you do not have, there is no second session.

How it is paid for

The app is free with ads and has an optional Pro subscription. That is not an ideological decision: AI plan generation and physique analysis call models that charge per use, and the server sits on a small instance that somebody has to pay for every month.

What is a decision is what sits behind the wall. Workout logging, the full library, the tutorials, the strength test and the premade routines are free and will stay free, because they are the part that lets someone train well without paying anything.

How it is built

JavaScript frontend with Vite and React, packaged for Android with Capacitor. Node and Express backend with SQLite on a small AWS instance behind Nginx. Physique analysis combines MediaPipe Pose for geometry with Gemini for vision and DeepSeek for reasoning over the numbers.

It is translated into fourteen languages, including all one hundred and thirty long tutorials. That part is by far the most time-consuming.

What is missing

iOS. It is built with Capacitor, so porting it is technically possible, but it takes time and a Mac, and right now the effort is going into making the Android version good.

If you find a bug, if a translation reads badly in your language, or if a tutorial says something wrong, write in. You will get an answer.