Turn on Chrome's built-in AI for AuditFocus

AuditFocus Pro and Agency use Chrome's built-in Gemini Nano model, entirely on your device, to suggest plain-English fixes for WCAG violations and generate alt text for flagged images. If Chrome's Prompt API is available on your machine, AuditFocus will light up those features automatically. This guide walks you through the one-time setup.

System requirements

Chrome's built-in AI runtime has hardware and OS constraints. Before you start, confirm your machine meets them. These minimums are set by Google for the Prompt API on device.

Chrome

Version 138 or later

Windows 10/11, macOS 13+, Linux

Storage

~22 GB free for the model

GPU / VRAM

4 GB VRAM or Apple Silicon

RAM

8 GB minimum, 16 GB recommended

Network

Unmetered connection for initial download

ChromeOS, Android, and iOS are not supported yet. If you're on one of those, AuditFocus still works for scanning and Markdown export on the free tier, but the AI fix suggestions and alt-text features will stay disabled until Chrome ships built-in AI to your platform.

Not sure what version of Chrome you have? Open chrome://version in a new tab. The number after "Google Chrome" on the first line should be 138 or higher.

Step 1 · Enable the Prompt API flag

Open Chrome flags

Copy this URL and paste it into the Chrome address bar (flag URLs can't be clicked as links for security reasons):

chrome://flags/#prompt-api-for-gemini-nano

Set the flag to Enabled

You'll see a flag titled Prompt API for Gemini Nano. From the dropdown on the right, choose Enabled.

While you're on this page, it's also worth enabling these related flags if you see them:

Enables optimization guide on device at chrome://flags/#optimization-guide-on-device-model set to Enabled BypassPerfRequirement
Summarization API for Gemini Nano if present, set to Enabled (optional, doesn't affect AuditFocus today)

Relaunch Chrome

After changing a flag, Chrome shows a blue Relaunch button at the bottom right. Click it. Your tabs will reopen after restart.

Step 2 · Download the Gemini Nano model

The Prompt API needs the on-device model component. Chrome downloads it in the background, but you can trigger and confirm the download through the Components page.

Open Chrome components

Paste this URL into the address bar:

chrome://components

Find Optimization Guide On Device Model

Scroll to the entry named Optimization Guide On Device Model. If the version shows as 0.0.0.0, the model hasn't been downloaded yet.

Click Check for update. The status will change to "Downloading" and then, once complete, the version will show something like 2024.11.xx.xxxx.

The model is roughly 2 GB compressed. First-time download can take several minutes on a fast connection, longer on metered ones. You only pay this cost once per Chrome profile.

Step 3 · Verify the setup works

Open the Chrome DevTools console on any regular webpage (right-click → Inspect → Console tab) and paste this check:

(async () => {
  if (!('LanguageModel' in self)) {
    console.log('Prompt API not available in this build');
    return;
  }
  const status = await LanguageModel.availability();
  console.log('LanguageModel availability:', status);
})();

You're looking for one of these responses in the console:

"available" — you're done, AuditFocus AI features are ready to go.
"downloadable" or "downloading" — the model still needs to finish downloading. Check back in a few minutes.
"unavailable" — something is off, see troubleshooting below.

When it works inside AuditFocus: open the side panel, run a scan on a page with at least one violation, expand a finding, and click Suggest fix or Generate alt text. You should see a streaming AI response with a code snippet or alt string you can copy.

Troubleshooting

Availability says "unavailable"

The most common causes:

Flag didn't stick. Revisit chrome://flags/#prompt-api-for-gemini-nano and confirm it reads Enabled, not Default.
Hardware doesn't meet the minimum. Chrome silently disables the API on machines under the VRAM/RAM floor. Try enabling chrome://flags/#optimization-guide-on-device-model as Enabled BypassPerfRequirement and relaunching.
Free disk space. You need at least 22 GB free for the model to download and expand.
Enterprise policy. If your Chrome is managed by an employer, the Prompt API may be disabled by policy. Check chrome://policy for GenAILocalFoundationalModelSettings.

Model version stays at 0.0.0.0

Chrome will not start the download until you've actually tried to use the Prompt API once in a page where the flag is enabled. Paste the verification snippet above into DevTools, leave the tab open for a minute, then revisit chrome://components and click Check for update again.

"Suggest fix" button in AuditFocus does nothing

If AuditFocus can't reach the Prompt API, the button falls back to a link pointing at the axe-core rule documentation. That's by design so the extension stays useful. To re-enable the AI pathway:

Confirm LanguageModel.availability() returns "available" using the snippet above.
Close and reopen the side panel so the panel re-checks feature detection on load.
If it still doesn't light up, email dev@loopheadlabs.com with your Chrome version (chrome://version) and OS. We'll dig in.

Does this work in Chromium, Edge, Brave, or Arc?

At launch, Google ships the Prompt API and the Gemini Nano model only in Chrome Stable and Chrome Canary. Chromium-based browsers that don't bundle the Optimization Guide component will not have the API available, so AuditFocus falls back to the docs-link mode on those browsers. The extension still runs fine there — scans, exports, and history all work. Only the on-device AI features require Chrome.

What stays on your device

Everything. AuditFocus never sends element snippets, scan results, or AI output to any server. The Prompt API runs inference locally using the Gemini Nano model that Chrome stored on your disk. We don't have an API key to give you, because there's no network call in the pipeline to make one on.

For the full breakdown of what we store and what we don't, see the Privacy Policy.