🧠All Things AI
🧠 All Things AI
Intermediate

Gemini Nano — On-Device AI

Gemini Nano is the smallest member of the Gemini model family, designed to run entirely on-device — on your Android phone or Chrome desktop browser — without ever sending data to Google's servers. It is the infrastructure behind many “instant” AI features that work with zero latency and no internet connection.

Why On-Device Matters

Running AI on the device itself instead of in the cloud offers four fundamental advantages:

  • Privacy: Your data never leaves the device. Input is processed locally and nothing is sent to external servers.
  • Latency: No network round-trip. Responses are near-instant, enabling real-time features like live transcription and instant Smart Reply.
  • Offline functionality: Works without an internet connection — critical for reliability in low-connectivity environments.
  • Cost: No per-call API cost to the device manufacturer or the user. Inference runs on the device's Neural Processing Unit (NPU).

Gemini Nano on Android

On Android, Gemini Nano is delivered via Android AICore — a system-level service that manages on-device model downloads and inference. The developer API became available experimentally in October 2024.

Current on-device features powered by Gemini Nano on Android include:

  • Smart Reply in Gboard: Context-aware reply suggestions that understand the full conversation, not just the last message.
  • Summarise in Recorder: Automatic transcription and summarisation of recorded audio — all processed on-device.
  • Scam call detection: Real-time analysis of phone call audio patterns to flag suspected scam calls mid-call on Pixel 9 series devices.
  • Notification summarisation: Condenses stacked notifications from apps into a single coherent summary.
  • Pixel AI features: Various Pixel 9-exclusive capabilities including on-device image captioning and contextual suggestions.

Gemini Nano in Chrome

Chrome 137 and later on desktop includes Gemini Nano via the Chrome Built-in AI APIs. This is currently a developer-facing API — end users benefit through features that use it, but direct API access is for developers building Chrome extensions and web apps.

Current features in Chrome powered by Gemini Nano:

  • Scam detection in Safe Browsing: On-device page content analysis to detect scam pages without sending page text to Google.
  • Help me write: In-browser writing assistance in the Chrome sidebar, powered by on-device inference.
  • Page summarisation: Summarise the current web page within Chrome's side panel.

Chrome for Android does not yet support the built-in Nano APIs — this is desktop-only for developers as of early 2026. In March 2026, Google expanded Gemini in Chrome to more countries and 50+ languages.

Developer Access

Developers can access Gemini Nano capabilities through two paths. On Android, the Android AICore API allows apps to perform on-device inference directly. On Chrome desktop, the Chrome Built-in AI APIs (available as an experimental origin trial) let web developers call Gemini Nano from JavaScript within a web page or extension. Google's intent is to eventually standardise on-device AI access through web platform APIs, making it model-agnostic and portable across browsers.

Checklist

  • What are the four key advantages of running AI on-device rather than in the cloud?
  • What is Android AICore, and what does it do on Android devices?
  • Which Chrome version introduced Gemini Nano built-in AI APIs?
  • Is Gemini Nano available in Chrome for Android as of early 2026?
  • Name two different features powered by Gemini Nano — one on Android and one in Chrome.