Description of the Change

Chrome is experimenting with adding built-in AI to their browser. The model they include is fairly lightweight and doesn't give quite as good results as larger models, but since it's built-in, there's no cost and all data stays private (also allows easy testing of some AI features without having to sign up for something).

At the moment, this feature is still being worked on and is only available in Chrome Canary, so there's a chance this never releases publicly or it may change quite a bit before the public release.

This PR integrates this browser AI API into our Title Generation, Excerpt Generation and Content Resizing Features. Because this API is 100% javascript, the way we integrate is a little unique compared to our other Providers:

• We still register a new Provider for Chrome AI
• This Provider still listens for REST requests on the same endpoint as our other Providers
• Instead of sending off API requests when a REST request comes in, we return the prompt and content
• I've added a new helper JS method that checks if the Chrome AI API is available
• This helper method is now used if the REST request returns the above data instead of the existing data structure. The AI request is then made at the javascript level, passing in the prompt and content returned from the REST request

How to test the Change

1. Install Chrome Canary v128.0.6545.0 or newer
2. Open a new tab in Chrome, go to chrome://flags/#optimization-guide-on-device-model and select Enabled BypassPerfRequirement
3. Go to chrome://flags/#prompt-api-for-gemini-nano and select Enabled
4. Relaunch Chrome
5. Open DevTools and send (await ai.languageModel.capabilities()).available; in the console
6. If this returns “readily”, then you are all set
7. If it fails, force Chrome to recognize that you want to use this API. To do so, open DevTools and send await ai.languageModel.create(); in the console. This will likely fail but it’s intended
8. Relaunch Chrome
9. Open a new tab in Chrome, go to chrome://components
10. Confirm that Gemini Nano is either available or is being downloaded. You'll want to see the Optimization Guide On Device Model present with a version greater or equal to 2024.5.21.1031. If there is no version listed, click on Check for update to force the download
11. Once the model has downloaded and has reached a version greater than shown above, open DevTools and send (await ai.languageModel.capabilities()).available; in the console. If this returns “readily”, then you are all set
12. Within the Title Generation, Excerpt Generation and Content Resizing Features, select Chrome AI as the Provider
13. Test that all three Features still work as expected. Note I had to massage the prompts a bit to get better results but as long as you're seeing results show up, things are working

Changelog Entry

Added - Chrome AI as a new text generation Provider, allowing completely on-device AI usage

Credits

Props @dkotter, @jeffpaul, @iamdharmesh, @faisal-alvi

Checklist:

• I agree to follow this project's Code of Conduct.
• I have updated the documentation accordingly.
• I have added Critical Flows, Test Cases, and/or End-to-End Tests to cover my change.
• All new and existing tests pass.