A new experimental handheld AI device developed by the public-interest initiative Current AI in collaboration with India’s government-backed Bhashini project suggests a different path for artificial intelligence—one that is open-source, portable, and capable of running offline. The prototype was demonstrated at India’s AI Impact Summit, where the small device used its camera and microphone to identify objects and answer questions in languages such as Hindi and English. It contains a touchscreen, microphone, camera, speaker, storage, and battery, allowing it to perform AI tasks directly on the device without relying on cloud servers.
The device is designed to support 22 Indic languages, reflecting a major effort to make AI more accessible for diverse linguistic communities in India. During the demonstration, the system scanned a table of candy bars and described what it saw, then answered follow-up questions conversationally. The project aims to show that AI systems can run effectively even on limited hardware, making them useful in regions with low or unreliable internet connectivity.
Developers plan to release the device’s hardware designs and software instructions as open source, allowing anyone to build, modify, or develop applications for it. The goal is to create a platform where developers can install different AI models and run them locally, enabling people to customize the device for specific needs such as translation, accessibility tools, or education. This approach could empower communities and researchers to experiment with AI without depending on large tech companies’ cloud infrastructure.
Supporters say projects like this represent the idea of “frugal AI”—smaller, affordable systems designed for practical use in diverse environments rather than massive data-center models. By focusing on open technology and localized language support, initiatives like Current AI and Bhashini hope to build alternatives to Big Tech–dominated AI ecosystems and ensure that more countries and communities can shape how AI develops.
