Developer Built Tiny AI Virtual Aquarium

A custom 14.3-million-parameter model now runs locally on a small touchscreen device to manage digital fish behavior.

Updated on Sept. 28, 2026 in Aquariums

Bold vector editorial illustration of a stylized fish swimming within geometric chip-like modules in coral and teal. — A developer has launched an ESP32-based virtual aquarium, using a 14.3-million-parameter model to simulate autonomous fish behavior offline on low-cost hardware. AI Illustration. Upload story photo >

Live Poll

Do you feel that modern smart home hobby projects are becoming too complex for you?

In September 2026, a developer launched an ESP32-based virtual aquarium that uses a compact language model to simulate fish activity. The system is designed to run entirely offline on low-cost hardware.

Why it matters

This project highlights how sophisticated AI workloads can be condensed to function on basic microcontroller platforms without cloud reliance. It represents a significant step in making local, autonomous digital environments accessible at a lower cost.

The system utilizes 14.3 million parameters in a 7.56 MB quantized file, delivering rendering speeds of 25-30 frames per second on a 1.8-inch display. The core hardware board typically costs $27.99, or Rs 3,099 at Indian distributors.

The players

ESP32-S3

A low-power microcontroller system often used for embedded AI projects and home automation devices.

The details

To fit the model onto 8 MB PSRAM, the developer used four-bit quantization and memory-mapped weights directly from flash. The model was trained via knowledge distillation using over 51,000 distinct fish scenarios to mimic a much larger teacher model. Users can flash the firmware and model onto the hardware by connecting the device to a computer running Chrome or Edge and utilizing ESP Web Tools.

The virtual-pet concept dates back to the 1990s.
The flee_shadow behavior was removed from the model in September 2026.

The Home Front

This project modernizes the 1990s Tamagotchi virtual-pet concept by replacing static programming with a locally running, learning language model. It follows the broader trend of bringing edge-computing capabilities into household devices.

If you are interested in experimenting with local AI at home, check the compatibility of your existing microcontroller boards with ESP Web Tools. You can monitor this developer project for future updates, such as findings on power efficiency and battery life.

The takeaway

This aquarium demonstrates that compact AI can provide autonomous, responsive digital experiences without requiring a constant internet connection. Owners of similar microcontroller devices may soon see more hobbyist projects that emphasize local data privacy and offline processing.