Google developed an AI translator that functions completely offline on a Raspberry Pi.

      Google, in collaboration with its Antigravity team, has created a prototype of an AI translator that operates entirely offline on a Raspberry Pi 5 utilizing the Gemma 4 E2B model. This translator functions without needing any internet connection. It features a microphone and speaker housed within a custom 3D-printed case, making it a fully portable translation device capable of working anywhere. The project is available as open source on GitHub.

      Gemma 4 was introduced by Google in April and comes in four model sizes, with the E2B variant specifically designed for compact hardware like the Raspberry Pi and Jetson Nano. The Translator prototype serves as a practical example of this technology. In contrast to most AI translation services, including Google Translate, which rely on cloud connectivity to process speech, this device’s on-device operation removes that necessity. This is particularly beneficial for travelers in areas with limited connectivity, organizations needing to manage sensitive discussions, and emergency responders working without infrastructure.

      This initiative is primarily a technology demonstration rather than a product launch. Google has not indicated plans to mass-market the device or incorporate it into any existing products. Nonetheless, it suggests a future possibility where AI assistants can function independently of data centers. Antigravity was revealed as Google’s agentic development platform at I/O, and the Translator is one of its initial projects aimed at non-developers—presenting a straightforward physical device that excels in a specific task without requiring an internet connection, all for under $100.

      As open models continue to become smaller yet more powerful, similar devices could eventually find their way into travel accessories, tools for accessibility, educational settings, or emergency response kits. The Raspberry Pi 5 retails for approximately $80, while the Gemma 4 E2B model is freely available under the Apache 2.0 license. The 3D-printed enclosure and associated code are also open source, making the necessary components for constructing a functional offline translator accessible to anyone interested in building one.