Recent versions of Chrome ship a small language model, Gemini Nano, and expose it to web pages through a family of experimental JavaScript APIs. This page detects what your browser supports and lets you try each one. Everything runs locally; no text leaves your device.
Each API is probed for presence and readiness. States follow the spec: available (ready now), downloadable / downloading (model fetch needed), or unavailable.
These APIs are currently a Chrome feature on desktop. To try them:
chrome://flags/#optimization-guide-on-device-model
and
chrome://flags/#prompt-api-for-gemini-nano, then relaunch.
| API | Global | Present | Availability |
|---|
The first run of any API may download the model. Progress is shown inline. Panels for unsupported APIs are disabled with a reason.
The built-in AI APIs expose on-device models as global classes such
as LanguageModel.
You call availability()
to check readiness, then create()
to start a session, optionally watching a download progress event.
These APIs are experimental and evolving across Chrome versions; some are gated behind origin trials or flags. They do not work in headless or server-side browsers, which lack the model and the hardware to run it.