Web Highlights Docs

Appearance

Troubleshooting Chrome AI Features ​

Chrome 138+ or Microsoft Edge 4.49.0 needed

Our AI features requires Chrome 138+ or Microsoft Edge 4.49.0. Please update your browser to use our AI summary feature and all other AI Features.

If you're experiencing issues with Web Highlights' AI summarization features in Chrome, this guide will help you resolve them. These features rely on Chrome's built-in AI capabilities, which can sometimes encounter compatibility issues.

Browser Requirements ​

Our AI features require your browser to have a local LLM (Large Language Model) installed. Chrome uses Google's Gemini Nano model, while Edge uses Microsoft's Phi-3 model - both run directly on your device.

Why is a download required? ​

The first time you use an AI feature, your browser downloads the model to your device. The download is about 3–4 GB — one time, and your browser manages it.

The 22 GB is free space, not the download

Chrome won't start the download unless you have about 22 GB of free disk space — but it does not use 22 GB. That number is headroom Chrome wants on the volume before it commits; the model itself is about 3–4 GB. (If your free space later drops below 10 GB, Chrome removes the model and re-downloads it once there's room again.)

Downloading the model to your device is what buys you:

  • Privacy: Your data stays local
  • Speed: No cloud processing delays
  • Offline: Works without internet once downloaded

Your content never leaves your device 🔒

This download is the privacy feature. Web Highlights has no AI server: your pages, highlights, notes, and questions are processed by the model inside your own browser — nothing is uploaded for AI, not even to us. You can verify it yourself in your browser's DevTools network tab. Once the model is there, AI even works offline.

Why our AI is privacy-first →

Chrome: Will automatically download Gemini Nano when you first try to use an AI feature Edge: Will automatically download Phi-3 when you first try to use an AI feature

You must keep the browser open during the download process for either browser.

Check Download Status

Monitor your download progress at:

  • Chrome: chrome://on-device-internals/ → "Model Status"
  • Edge: edge://on-device-internals/ → "Model Status"

Browser Restart Required

A browser restart may be required after the download completes for the AI features to become available.

System Requirements ​

Experimental Features ⚠

Chrome's AI features are experimental and may have compatibility issues on certain systems. The solutions in this guide address the most common problems users encounter.

Chrome's AI features require specific hardware and software:

Operating System ​

  • ✅ Windows 10/11
  • ✅ macOS 13+ (Ventura and later)
  • ✅ Linux
  • ❌ Chrome for Android, iOS, or ChromeOS

Hardware Requirements ​

  • Storage: About 22GB free space — Chrome's threshold before it downloads; the model itself is ~3–4 GB
  • GPU: More than 4GB VRAM
  • Network: Unlimited/unmetered connection

Common Issues ​

AI features don't work on insecure (http://) pages ​

Chrome only exposes its built-in AI on secure pages (https://). On a plain http:// page the on-device model is not available at all — even in a fully up-to-date Chrome with the model installed. Your browser is fine; the page is the blocker.

What you can do:

  • Reopen the page over https:// if the site supports it
  • Open your highlights in your Web Highlights dashboard — it always runs on a secure page, so AI summaries, chat, and flashcard generation work there

This is a Chrome platform restriction (the AI APIs require a secure context), not a Web Highlights limitation. Curious why our AI runs on your device in the first place? See why our AI is privacy-first.

"InvalidStateError: The device is unable to create a session to run the model" ​

This error typically occurs when Chrome's AI model (Gemini Nano) is not properly installed or configured.

"The AI model is unavailable. Please try again later." ​

This indicates that the AI model is not ready to use.

Quick Fixes ​

1. Enable Chrome AI Flags (Most Common Solution) ​

  1. Open Chrome and navigate to chrome://flags/
  2. Search for and enable these flags:
    • #optimization-guide-on-device-model → Set to: Enabled BypassPerfRequirement
    • Prompt API → Set to: Enabled
    • Summarization API for Gemini Nano → Set to: Enabled
    • #enable-on-device-model → Set to: Enabled
  3. Click "Relaunch" when prompted
  4. Try the AI feature again

2. Check Chrome Version ​

Ensure you're using Chrome version 138 or later:

  1. Go to chrome://settings/help
  2. Update Chrome if available
  3. Restart Chrome

Apple Silicon Users ​

If you're using a Mac with Apple Silicon (M1, M2, M3), you may need to use the bypass flags mentioned above, as Chrome sometimes has difficulty detecting VRAM on these systems.

Advanced Troubleshooting ​

Check AI Model Status ​

  1. Go to chrome://on-device-internals
  2. Navigate to the Model Status tab
  3. Look for any error messages or status information

Check Optimization Guide status ​

  1. Go to: chrome//components
  2. Navigate to Optimization Guide On Device Model
  3. Click check for updates

Monitor Model Download ​

If the model is downloading:

  1. Don't close Chrome during the download
  2. Keep the tab active where you triggered the download
  3. Monitor progress at chrome://on-device-internals → Model Status
  4. Expected download time: 5-30 minutes depending on your internet connection

Test AI Model Availability ​

Open Chrome DevTools (F12) and run this command in the console:

javascript
await LanguageModel.availability();

Expected responses:

  • "unavailable" → Check system requirements and flags
  • "downloadable" → Ready to download (good!)
  • "downloading" → Model is currently downloading
  • "available" → Model is ready to use

Alternative: Use Chrome Canary ​

Chrome Canary often has better support for experimental features:

  1. Download Chrome Canary
  2. Install alongside your regular Chrome
  3. Apply the same AI flags
  4. Test the AI features there

Alternative: Use Microsoft Edge ​

Microsoft Edge uses its own AI models and doesn't rely on Chrome's Gemini Nano download system, which is why it continues to work while Chrome may have compatibility issues.

If you have trouble using Chrome's AI, install our add-on on Edge.

Complete Reset (If Nothing Else Works) ​

Remove Chrome Completely ​

If you continue to experience issues, you may need to completely remove Chrome and reinstall:

  1. Close Chrome completely
  2. Remove Chrome applications:
    bash
    sudo rm -rf "/Applications/Google Chrome.app"
sudo rm -rf "/Applications/Google Chrome Canary.app"
  3. Remove Chrome data:
    bash
    rm -rf ~/Library/Application\ Support/Google/Chrome*
rm -rf ~/Library/Caches/Google/Chrome*
rm -rf ~/Library/Preferences/com.google.Chrome*
  4. Restart your computer
  5. Download and install Chrome fresh from google.com/chrome
  6. Enable the AI flags before trying AI features

Prevention Tips ​

  • Keep Chrome updated to the latest version
  • Ensure adequate storage space (~22GB free — Chrome deletes the model if free space later drops below 10GB)
  • Don't interrupt Chrome during AI model downloads
  • Monitor your GPU VRAM if using Windows/Linux
  • Use stable internet connections for model downloads

Still Having Issues? ​

If none of these solutions work:

  1. Check the Chrome AI documentation at developer.chrome.com/docs/ai/get-started
  2. File a Chrome bug report at Chrome Support