WebAI lets websites and extensions use AI models running on your computer, without sending page content to a cloud AI service.
@Opera built the inference engine right into the browser. Developers get three APIs for text generation, classification, and finding related content.
Opera also made six example extensions, including one that quizzes you on an article and another that highlights paragraphs relevant to your project. These small tools make total sense to run locally.
WebAI is now an opt-in early preview in Opera One. It supports local models including Gemma and Apple's on-device foundation model on Macs.
The three APIs are what caught my eye, especially the related-content one, since that is the one where a page would need the most context. When a site asks for related content, does it get back passages it can highlight directly, or does it have to re-find them in the page itself? And if the user only has one local model installed, does the site get to say which capability it needs, or does WebAI just offer whatever model is there?
Report
A very nice and extra useful addition. Hopefully, all security and data privacy measures are in place properly.
How does a website negotiate which local model it gets to use? Wondering what stops a random site from quietly spinning up the biggest model available and eating my battery for breakfast.
Replies
Hi everyone!
WebAI lets websites and extensions use AI models running on your computer, without sending page content to a cloud AI service.
@Opera built the inference engine right into the browser. Developers get three APIs for text generation, classification, and finding related content.
Opera also made six example extensions, including one that quizzes you on an article and another that highlights paragraphs relevant to your project. These small tools make total sense to run locally.
WebAI is now an opt-in early preview in Opera One. It supports local models including Gemma and Apple's on-device foundation model on Macs.
classification and related content seem like sensible first uses for a model already in the browser
i would want the site permissions to make it clear which page can use it and what context that page gets
Love to see Opera embracing local models first on this, super cool and definitely the right direction IMO
The three APIs are what caught my eye, especially the related-content one, since that is the one where a page would need the most context. When a site asks for related content, does it get back passages it can highlight directly, or does it have to re-find them in the page itself? And if the user only has one local model installed, does the site get to say which capability it needs, or does WebAI just offer whatever model is there?
A very nice and extra useful addition. Hopefully, all security and data privacy measures are in place properly.
How does a website negotiate which local model it gets to use? Wondering what stops a random site from quietly spinning up the biggest model available and eating my battery for breakfast.