Opera announced WebAI on October 27, a browser feature and proposed web standard that lets websites and extensions run AI models on the user’s own computer instead of sending requests to a cloud AI provider. The company put it into an early preview that users must switch on in settings, and it described the arrangement as one in which page content does not have to leave the device.
Catch up on AI in 3 minutes a day: 700,000+ readers get The Neuron, a daily email with the latest AI tools, tutorials and news.
We may earn a commission if you sign up.
The pitch comes from Krystian Kolondra, Opera’s executive vice president for browsers and AI. “A lot of what people want from AI doesn’t need a data center,” he said in the company’s announcement, adding that users “will always decide what runs on” their machines. Those are the vendor’s claims, and the details below separate what Opera has documented from what it has asserted.
Three APIs and the models behind them
In its announcement post, Opera introduces three interfaces: Prompt, for generating text; Decision, for classifying information; and Semantic Embeddings, for finding relationships between pieces of content. A developer calls them from page or extension code, and the browser hands the work to a model installed on the machine. Opera’s separate WebAI playground page lists a longer set of eight on-device APIs, including a summarizer, rewriter, translator and proofreader, so the three in the post look like the core of a wider toolkit.
Which models sit underneath depends on who is reporting. Opera says developers can use open-weight models supplied through its own inference engine, along with compatible models already present on the device. TechCrunch named Google’s Gemma as an example and added that on Macs the feature also reaches Apple’s on-device foundation model, a pairing 9to5Mac reported the same day. Opera calls the design model-agnostic, so no single provider is meant to be required.
What runs on the device, and what the record leaves open
Opera’s account is that inference happens inside the browser, with no separate app, local server or API key. Because processing stays on the machine, the company says, page content need not reach a cloud AI provider and developers avoid paying per request. TechCrunch quoted the announcement’s own wording that users “don’t need AI subscriptions” and that the information the local AI handles stays on their own device. No independent audit of those privacy statements was found. They describe a design intention in a preview product, and the sources read for this report did not publish a network-traffic analysis or a third-party review.
The same sources leave a few things unstated. A website that calls WebAI is still a website, and it still talks to its own servers for everything else it does. Opera’s post says nothing is downloaded until a user opts in, which implies models are fetched at some point, but the post as read does not say from where or how large the files are. Opera adds that users can decide which sites may reach local models and can turn the feature off at any time.
The shift also changes who pays. Opera’s stated pitch to developers is that they can add AI features without operating AI infrastructure or paying for each request, because the user’s processor does the work and the site owner does not rent inference capacity. The cost does not vanish; it lands on the visitor’s battery, memory and patience. A site that leans on a local model must also plan for visitors whose machines cannot run one, including every Intel Mac and Linux desktop, and for output that will vary with whichever model a given person has installed.
Who can run it today
The hardware list is narrow. According to Opera, WebAI works on 64-bit Windows PCs with Intel or AMD processors and on Macs with Apple silicon. It is not available on Intel Macs, Linux or Windows on Arm. Opera also acknowledges that response speed depends on the device, which matters for a feature that moves the compute burden from a data center onto a laptop. TechCrunch put the same trade-off in plain terms: local models demand more of the user’s computer.
The rollout description differs between outlets. Opera’s post describes the early preview as covering Opera One, Opera GX and Opera Neon, while TechCrunch reported that Opera One has it now and that the gaming browser GX and the AI-agent browser Neon will “soon” follow, without a date. 9to5Mac likewise called One the first to offer it. Readers checking a specific install should treat Opera’s own release notes as the authoritative list.
To show what developers might build, Opera released six experimental extensions: Project Lens, Got It?, Questionable Intelligence, Semantic Art, RecordIn Alchemy and CIAO. The company labels them examples rather than built-in features. 9to5Mac listed everyday uses such as sorting tabs, flagging clickbait and spotting relevant job posts, all tasks where a small model on the laptop could stand in for a round trip to a hosted one.
From a 2024 experiment to a proposed standard
Opera has been here before. In April 2024 it announced Opera One Developer as the first browser with built-in local language models, a developer-channel test that let users download and run models on their own machines. WebAI turns that browser-side capability into something websites can call, which is why Opera frames it as an open interface rather than an Opera-only tool.
That framing carries the open question. Opera describes WebAI as a proposed web standard and says Google and Microsoft are among its backers, according to the announcement as summarized for this report. A proposal from one browser maker is not an adopted standard, and no formal specification or source repository was linked from the announcement. Whether other browsers ship the same interfaces, and whether sites bother to use them, will determine if local AI in the browser becomes a common layer of the web or remains one vendor’s preview.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
Catch up on AI in 3 minutes a day: Join The Neuron →
We may earn a commission if you sign up.
More from Morning Overview
- Seven of Earth’s nine planetary boundaries are breached, and all seven are worsening
- The NSA says three phone features should be off whenever you aren’t using them
- Hurricane Hunter radar shows four warning signs that a tropical cyclone is about to strengthen, a University of Miami study found
- Tropical Storm Rachel is dumping up to 12 inches on four Mexican states on its way to major hurricane strength
