An AI translator that works without internet: Google's device that anyone can build

Artificial intelligence continues to open up new possibilities and, this time, a team of Google engineers has developed a portable translator capable of working without an internet connection, with a design that can be replicated by anyone interested in building their own version.

The project is based on a Raspberry Pi 5 and uses the AI model Gemma 4 E4B, an open-weight multimodal model optimised to run efficiently on local devices. Thanks to this approach, the translator can perform real-time translations without relying on the cloud or requiring a high-performance computer.

The development was carried out using Google Antigravity, the company's artificial intelligence software experimentation platform. In addition to presenting the device, its creators have published the entire project on GitHub, allowing any user to download the code, compile it and create their own version.

According to the videos shared by Google Antigravity, the translator features a 3D-printed casing made up of two stacked modules. Inside, it houses the Raspberry Pi 5, while the exterior includes a screen to display translations, a microphone to capture speech and a speaker to play back the translated result.

Despite its compact size — the Raspberry Pi 5 is similar in dimensions to a credit card — this small computer provides enough power to run applications related to programming, robotics, automation and artificial intelligence, making it an ideal foundation for this translation device.

Its external design resembles a small robot, while its operation has been simplified as much as possible. Users only have one button to start a conversation and a rotary selector that allows them to change the translation language instantly.

From a technical perspective, the system relies on LiteRT-LM, a lightweight runtime environment specifically designed to run generative language models on devices with limited resources. Voice generation is handled by Moonshine Text-to-Speech, the technology responsible for converting translated text into audio played through the integrated speaker.

With this open-source project, Google shows that it is possible to create smart, portable and fully autonomous translators, eliminating the need for an internet connection and allowing any developer or technology enthusiast to reproduce the device using relatively affordable components.