Gymnasia

What makes it different

There are plenty of gym apps, and plenty of diet apps. What changes here is what it carries inside: an AI agent that acts as your personal trainer.

And it is not a chat bolted onto the side of the app to answer the odd question. It is an agent with real access to your data: you say "build me a three-day push routine" or "log that I had two eggs and a slice of toast" and it does it, searching the exercise library, writing into today's diet, reading your measurements to decide, and remembering what you told it last week.

The second thing that makes it different: that agent runs entirely on your phone. There is no server of mine your data passes through, because there is no server. The app calls the AI provider directly with a key you supply, and everything else, deciding what to do, running it and storing it, happens inside the phone. Your weight, your diet and your routines never leave it.

The conversational coach

This is the agent you talk to. It operates the app through a catalogue of tools: it searches exercises and foods, reads and writes your diet, creates routines, logs measurements and stores what you tell it about yourself so it is there in the next conversation. What it is able to do is exactly that list, and nothing more.

It is BYOK (bring your own key): you supply the API key and it is stored on the device, so you choose which provider to use and control what you spend. Its system prompt is fetched remotely with an embedded fallback copy, so it can be fixed without shipping a new release and the app keeps working offline.

The food estimator

The second agent, independent from the first one: you take a photo of the plate and it gives you calories and macronutrients. If there is a barcode in the photo it reads it and queries OpenFoodFacts, so for supermarket products the numbers are the real ones instead of a guess. Nothing is saved into your diet until you approve it.

Everything else

• Workout Routines

  Create custom routines with exercises, sets, reps and weights. Each routine can be categorized (strength, cardio, flexibility, etc.) and customized with icons. During training, a rest timer with sound alert notifies you when to continue.

• Exercise Library

  Open-source exercise repository with AI-generated images showing correct execution, in both male and female versions. Exercises include muscle group, required equipment, difficulty level and detailed instructions. Users can also create custom exercises.

• Diet tracking

  Daily meal logging with calorie and macronutrient calculation (protein, carbs and fats), personalized nutritional goals and barcode scanning against OpenFoodFacts.

• Body Measurements

  Track weight and other body measurements with historical data to visualize your progress.

Architecture

The app has no backend: both agents run entirely on the phone, so the user's data never leaves the device. The project is an npm workspace monorepo containing:

• apps/mobile: Expo React Native app (primary surface)
• packages/shared: Shared design tokens and TypeScript types
• ejercicios/: Open-source exercise repository (JSON + AI-generated images)

The mobile app uses AsyncStorage for local persistence and Expo SecureStore for API keys. Exercise images are served from the GitHub repository and cached on-device.

Technologies

• Mobile: React Native, Expo SDK 54, TypeScript
• AI: OpenAI, Anthropic and Google Gemini through adapters, with a BYOK scheme
• Nutrition data: OpenFoodFacts for barcode scanning
• Image Generation: Generative AI for exercise illustrations
• Build: EAS Build for Android APK generation

Source Code

The project is open source and available on GitHub.

Support

If you like the project, consider giving it a star on GitHub.

More about the agents

If you want more information about how the app is made, the two graphs above explained step by step, the tool loop, the provider choice and why the system prompt lives remotely, it is all in Gymnasia: architecture of an AI agent that runs on the phone.