Google Brings Gemini Models to Android System Services for On-Device Assistants

AI · 6 min read

In a platform-level update, Google now permits OEMs to bundle optimized Gemini Nano and Lite variants with Android, enabling assistant features like offline transcription, smart replies, and local summarization.

The system service API exposes hooks for app developers to call the local model for short context tasks while deferring heavy generation to cloud Gemini Ultra when necessary. Privacy controls allow users to restrict what data streams are accessible to local models.

Google also introduced a model update framework that patches on-device weights over differential updates, reducing download sizes for security and capability patches. The move represents Google's blended approach to balancing privacy, performance, and compute costs.