Games

Kaurix

Find creatures with your camera, cook, and make them grow. Earn cards and fill the album.

Available in: Spanish · English · Portuguese · Italian

How it looks

Pick a thumbnail to move through the screens, or tap the large screenshot to enlarge it.

1 / 14

Kaurix creatures screen: the cycle's creatures, the ones found in colour and the missing ones as silhouettes

About Kaurix

Kaurix starts with an egg that isn't on the screen: it's in the room. The game tells you a creature is waiting — it shows you its silhouette, not where it is — and from there it's on you: you raise your phone and turn slowly until it appears, floating over the table, the shelf or the yard.

Finding it isn't enough: you have to hold it. It comes closer while you keep it in frame, and backs away if it slips out. Once it's close, you tap it. Sometimes the shell takes a couple of taps before it breaks.

Whatever hatches from that egg has to be fed, and that takes ingredients. They turn up while you look around — seventy-four of them, from fruit and vegetables to stones and flowers — and you gather them by tapping.

Then you cook: fifty-two recipes across three levels. Each one takes exactly three different things, but from the second level on, one of the three is a dish you already cooked, so meals combine with each other into better ones. Nothing to guess: the bag holds the recipes and tells you where each ingredient comes from.

Every creature that reaches adulthood gives you an illustrated card. The album has eight pages — one per creature, with its eight actions and its legendary in the middle — and all eight fill at once, so any growth can open a page you hadn't touched. The legendary is never drawn: it's what you earn by completing the eight on its page, and completing all eight pages finishes the game.

It runs on any Android with a camera. It deliberately avoids certified augmented reality: that leaves out almost every mid-range phone, and this game had to work on the phone you already own. No account, no connection needed to play, and your game lives on your phone.

Look around. Something moves.

Augmented reality wherever you are

Creatures and ingredients appear over whatever the camera is pointing at. The game tells you something is there, but not which way: you turn until you find it, and it comes closer while you hold it in frame.

Cooking is half the game

Every recipe takes three different things, and from the second level on, one is a dish you already cooked. Fifty-two between meals and potions, all of them listed in the bag.

From egg to adult

Each creature grows in stages from what you feed it. Filling the bar isn't enough: a potion is what unlocks the jump to the next stage.

An album that fills up

Eight pages of nine cards that fill at the same time. Completing a page's eight actions earns you its legendary; completing all eight pages finishes the game.

Tech specs

Kaurix brings augmented reality and computer vision together on the phone: an image recognition model trained for this game, running offline, and augmented reality designed to work on any Android with a camera, not just high-end devices.

React Native + Expo + TypeScript
Custom image recognition model
Dataset curated with CLIP
Native Kotlin module + ML Kit
Augmented reality without ARCore
3D art exported to animated WebP
Local save
i18n
Google AdMob + UMP
EAS Build

App: React Native + Expo + TypeScript
Android game with React Navigation and state with Zustand.
i18n
Spanish, English, Italian and Portuguese.
Architecture: 3D art exported to animated WebP
3D is used to produce the art, not at runtime: own Node tools (ffmpeg, sharp) turn it into animated WebP and sprite sheets, and a config plugin enables animated WebP on Android.
Native & device: Native Kotlin module + ML Kit
Own module built with the Expo Modules API that loads the .tflite model with ML Kit Image Labeling and, if it's missing, falls back to ML Kit's base model. It's complemented by an analysis of the scene's color.
Augmented reality without ARCore
Creatures are anchored to a direction in the world using the gyroscope (expo-sensors) over the live camera (expo-camera). On devices without a gyroscope the app picks another engine by itself. ARCore was ruled out on purpose because it leaves out most mid- and low-end phones.
AI & machine learning: Custom image recognition model
MobileNetV3 classifier trained with TensorFlow/Keras through transfer learning and fine-tuning: 48 ingredient classes, exported to TensorFlow Lite (3.3 MB) and run on the phone, with no images sent to any server.
Dataset curated with CLIP
Own Python pipeline: photos from Open Images v7 and iNaturalist (with license and author recorded), filtered with CLIP on PyTorch to keep only those showing the class, and a separate evaluation with real photos taken with the phone.
Data: Local save
All progress lives on the phone (AsyncStorage and expo-file-system), with no account or server.
Monetization: Google AdMob + UMP
Rewarded videos inside the game and consent for the European Union, the United Kingdom and Switzerland.
Tooling & delivery: EAS Build
Production builds for Google Play, with per-environment variables.