Looking for Chrome's built-in AI.
Chrome needs your click to start the download. It is a few gigabytes and happens once. Keep this tab open until it finishes.
Gemini Nano only runs in Chrome on Windows, macOS or Linux (and Chromebook Plus). It does not work on phones or in other browsers.
Chrome does not let web pages link to its own chrome:// pages. Press Copy, open a new tab, paste into the address bar and press Enter.
chrome://flags/#prompt-api-for-gemini-nanoset to Enabledchrome://flags/#optimization-guide-on-device-modelset to Enabled BypassPerfRequirementThen click the Relaunch button at the bottom of the flags page. If you do not see the second flag, finish step 3 first and come back.
Chrome needs at least 22 GB of free disk space on the drive that holds your Chrome profile, or it will not fetch the model.
chrome://componentsOptimization Guide On Device Model, then Check for updateIf the component is missing or shows version 0.0.0.0, open this page, press F12, choose Console, paste the line below and press Enter. It makes Chrome start the download.
await (window.LanguageModel ? LanguageModel.availability() : ai.languageModel.capabilities())Put all six files in one folder. Any folder name works.
Easiest: double-click index.html. If Chrome still says the API is missing, use localhost instead, which is the most reliable: open a terminal inside that folder and run one of these, then visit the address.
python -m http.server 8000Windows: py -m http.server 8000http://localhost:8000This page also re-checks by itself when you switch back to this tab.
Languages: Gemini Nano officially supports English, Spanish and Japanese. Russian and Polish may give poor answers or errors.
Works offline: after the one-time download the model runs completely on your device. You can turn off Wi-Fi and keep chatting. Nothing you type is sent anywhere.