Opera's new WebAI preview lets websites and extensions run AI models locally inside the browser. The page you are reading could soon do its own thinking on your hardware, with no cloud call, no API key, and no token bill.
Every AI feature on the web today phones home. You type into a page, the page ships your words to a data centre, and a model you will never see sends an answer back. Opera wants to cut the phone line. Its new WebAI, now in early preview across Opera One, Opera GX, and Opera Neon, lets websites and extensions run AI models locally, inside the browser, on the machine already sitting on your desk.
Opera has form here. In 2024 it became the first major browser to offer local AI models, in an experimental developer build. WebAI is that experiment grown up and opened to the whole web. Here is Opera’s own walkthrough, then the substance.
Opera demonstrates WebAI running inside the browser. Video: Opera
What WebAI actually is
WebAI is a proposed web standard, one Opera says carries backing from Google and Microsoft. The idea is simple to state and large in consequence: a site or extension asks the browser for AI, and the browser runs a model locally to answer. There is no separate AI app to install, no local server to configure, and no API key to hand over. The inference engine lives in the browser itself.
The models are flexible. Developers can use open weight models supplied through Opera, or compatible models already stored on your device, so nobody is locked to a single provider. And because the work happens on your hardware, the content being processed never has to leave for a cloud AI service.
Control sits with the user, at least on paper. WebAI is opt in. Nothing downloads until you switch it on in Settings, and you can switch it off again whenever you like. Opera executive Krystian Kolondra put the pitch plainly: “A lot of what people want from AI doesn’t need a data center. Your laptop can already do it, without sending your data anywhere.”
The tools developers get
Three interfaces anchor the preview. The Prompt API generates text. The Decision API classifies information. The Semantic Embeddings API finds relationships between pieces of content, the plumbing behind search and recommendation. Together they cover the jobs most site features actually need, and none of them requires the developer to rent GPU time.
For independent developers and small teams, the economics are the story. Local inference means no per request cloud costs and no AI infrastructure to run or secure. Your users get privacy as a side effect, because their page data can stay on their device. Opera also points at a quieter shift: anyone using AI coding tools can now build personal extensions that would never justify a cloud bill, tuned to one person’s habits and nobody else’s business.
Six demos, varying sanity
To show the shape of the thing, Opera built six experimental extensions. Project Lens colours the paragraphs of an article by how much each one matters to a project you describe. Got It? writes a short quiz about the article you just read. Questionable Intelligence asks playful questions about the page and charts the verdict. Semantic Art turns an article into moving art, with similar sentences drifting together. RecordIn Alchemy converts any article into an insufferably inspirational social media post, which is either a warning or a gift. CΒ·IΒ·AΒ·O puts two ASCII cats on stage and has them improvise a scene on any topic you name.
The cats are a joke. The pattern underneath them is not. Every one of these demos does its thinking on the reader’s machine, about the page the reader is already on, and none of them needed a server to do it.
How to try it, and the catch
Download or update Opera One, open Settings, then Opera AI, and turn on Enable Web AI API. That downloads a compatible local model, which can take a few minutes. Install one of the demo extensions and open it from the extensions menu.
Now the fine print. The preview runs on 64 bit Windows PCs with Intel or AMD chips and on Macs with Apple silicon. Intel Macs, Linux, and Windows on Arm are all excluded, which locks out a large share of the machines developers themselves use. Speed depends entirely on your hardware, and Opera concedes that local AI performance is still improving and still being made lighter. This is a preview in the honest sense: real, working, and unfinished.
The take
The browser is becoming the place AI lives, and the fight is over whose computer does the work. Cloud AI centralised intelligence inside a handful of companies and billed everyone by the token. WebAI proposes the opposite shape: a standard any site can call, models the user chooses, and processing that stays in the room. That it arrives as a proposed standard with Google and Microsoft behind it matters more than Opera shipping first. Standards are how experiments turn into defaults.
Whether it holds will depend on the models getting smaller and the hardware getting faster, both of which are already happening. The direction, though, is set. The smartest page on your screen may soon be the one that never asked a data centre for help.
Sources
Opera, “WebAI: Local AI for the web, now in Opera,” Opera News blog, October 2026. Developer documentation and the full API list live at next.opera.com/web-ai.
END OF TRANSMISSION
Stay curious. Question everything.
