BharatGPT-mini ran on a laptop—no internet, no cloud. CoRover.ai showed off BharatGPT-mini at Intel's AI PC Innovation Event in Conrad Bengaluru on April 24. The small language model ran as a conversational AI agent on Intel's Core Ultra Series 3 processor, handling text, voice, image, and video in multiple Indian and international languages—all without an internet connection.
The model processes everything on-device using Intel's OpenVINO toolkit, which optimises inference across the chip's CPU, NPU, and accelerators.
Who actually needs offline AI
Patchy connectivity is one reason to care. Defence, healthcare, and banking are another—sectors where sending data to an external server isn't always an option, legally or practically. On-device processing keeps everything local.
CoRover CEO Ankush Sabharwal, speaking at the event, described a three-tier architecture: on-device for speed and privacy, on-premise for control, and cloud for scale. His point was that each layer serves a different purpose, and on-device handles the cases where the other two simply aren't available.
CoRover already has deployments across government, BFSI, education, and defence verticals. The Intel demo extends that into individual PCs—field offices, classrooms, and desks where connectivity is unreliable or data sensitivity is high.
The broader pitch from Sabharwal: cloud AI made the technology accessible, but on-device is what makes it usable in the places that have so far been left out.