For the past few years, "adding AI to your website" has meant the same recipe: get an API key, send user data to someone's cloud, and watch the invoice. Opera would like to tear up that recipe and run the model on the visitor's own machine.
A Model in Every Browser Tab
On October 27, Opera launched WebAI as an opt-in early preview in Opera One. It lets websites and browser extensions use AI models that run locally in the browser. Opera says that means "no separate AI app to install, no local server to configure and no API key to get." At launch it taps models including Google's Gemma and, on Macs, Apple's on-device foundation model.
Developers get three APIs: Prompt for generating text, Decision for classifying information, and Semantic Embeddings for finding relationships between content. It runs on 64-bit Windows PCs and Apple silicon Macs, but not Intel Macs, Linux, or Windows on Arm, at least for now. Users turn it on under Settings, which downloads a local model. Site-specific permissions control which websites can use it. Opera also shipped six example extensions to show developers what's possible.
Privacy as a Feature, Not a Footnote
The pitch that matters is in Opera's own words: because the AI runs on your computer, "the content it processes doesn't need to be sent to a cloud AI provider." For anyone building sites in healthcare, legal, finance, or anywhere a compliance officer lives, that's a big change. A form that summarizes a user's notes without the notes ever leaving their laptop is a much easier conversation with your privacy team.
Now for the asterisk. Opera calls WebAI a proposed web standard "backed by companies such as Google and Microsoft," but its announcements don't name a standards body or a published spec, and right now Opera is the one shipping it. Building a public-facing feature that only works in one browser is a decision teams usually come to regret. The sensible move today is to experiment and use progressive enhancement: run AI locally when it's available, and fall back gracefully when it isn't.
There's also a quieter economic angle. When inference runs on the visitor's hardware, the per-request cloud bill disappears. That could make small AI features, like smarter search, auto-tagging, or content suggestions, viable for sites that could never justify a monthly API spend. It won't replace frontier cloud models for heavy lifting, but plenty of everyday website tasks don't need a frontier model.
The browser wars are back, and this time the prize isn't rendering speed but who gets to host the model. Bring popcorn, and a fallback plan.
If on-device AI has you rethinking which features your site could add without sending a single byte to the cloud, let's sketch out what that could look like for you.
Source: Opera Blog