Local Browser LLM

Enable Chrome built-in AI with Gemini Nano

Use this guide to turn on Chrome's local language model, verify that the Prompt API is ready, and diagnose download or hardware issues before using OmniFlows with browser-native AI.

01

Enable the required Chrome flags

Open a new Chrome tab and configure the two experimental features that allow Chrome to load and expose the local Gemini Nano model.

  1. Open chrome://flags/#optimization-guide-on-device-model and set it to Enabled. This enables Chrome's device-side optimization guide model, the core engine for loading and running local machine learning models.
  2. Open chrome://flags/#prompt-api-for-gemini-nano and set it to Enabled, or Enabled multilingual if you need multilingual input and output. This exposes Gemini Nano through JavaScript APIs such as LanguageModel.
  3. Click Relaunch in the lower-right corner of the flags page so both changes take effect.
02

Verify model availability in DevTools

After Chrome restarts, open any webpage, launch DevTools with F12, and run the availability check in the Console.

await LanguageModel.availability();
  1. available means the model is ready and can be used immediately.
  2. downloadable or downloading means Chrome supports the feature, but the local model is not ready yet.
  3. If the result is downloadable, interact with the page first, such as clicking or pressing a key, then call await LanguageModel.create();. Chrome requires user activation before it starts the model download.
03

Use Chrome's internal diagnostics page

If the model fails to initialize, reports an error, or appears stuck while downloading, open the device-side AI diagnostics page.

  1. Visit chrome://on-device-internals.
  2. Check the current Gemini Nano model version, download progress, and initialization state.
  3. Open the Model Status tab if loading fails. Chrome shows detailed error codes there, which are useful when diagnosing local model problems.
04

Confirm hardware and system requirements

If availability() returns unavailable, or the download keeps failing, check the hard requirements below.

  1. Storage: the drive that contains the Chrome user profile must have at least 22 GB free. If free space drops below 10 GB after download, Chrome may automatically remove the model.
  2. CPU mode: the machine needs at least 16 GB RAM and a processor with 4 or more cores.
  3. GPU mode: VRAM must be strictly greater than 4 GB. Prompt API flows that include audio input require GPU support.
  4. Network: the first download must happen on an unmetered connection, such as regular Wi-Fi or wired Ethernet. Chrome may block downloads on mobile hotspots or other metered networks to protect user data usage.

For more information, see: https://developer.chrome.com/docs/ai/get-started