Opera introduces a new web standard to let users deploy AI locally in their browser

Opera has introduced WebAI, a proposed web standard that lets AI models run directly in the browser on the user's own machine rather than in the cloud. It isn't tied to one model. Opera mentions Google's open-weight Gemma as an example, and on Macs it also supports Apple's on-device foundation model.

It's an opt-in early preview in Opera One for now, with Opera GX and the agent-focused Opera Neon coming next. Opera's pitch is simple: developers can add AI features without running infrastructure or paying per request, and users get AI without a subscription and without their data leaving the device.

We like the direction. A lot of everyday AI work, like classifying text, summarizing a page or spotting how pieces of information relate, doesn't need a frontier model in a distant data center. Running it locally is cheaper, keeps working when a service changes its pricing, and keeps personal data on the device. For people who have been pouring sensitive details into cloud chatbots without knowing where that data goes, this is a real improvement.

The word that matters is "standard." A feature in one browser is a product. A standard only matters if other browser makers adopt it, and Opera has a small share of the market. If Chrome, Safari and Firefox ignore WebAI or build their own incompatible versions, developers will have to write for several of them or skip local AI altogether. I'm watching for any sign of uptake beyond Opera.

The trade-offs are real. Local models put load on the user's hardware, and Opera itself says performance depends on the device. That splits the experience: people with newer machines get fast, private AI, while people on older or cheaper hardware get slow or degraded features. A web that assumes local AI could quietly become less usable for exactly the people who can least afford upgrades.

Local also doesn't mean risk-free. A model running in the browser can still be fed malicious page content, still make mistakes, and still be used by a site in ways the user didn't expect. Keeping data on the device is a meaningful privacy gain, but it isn't a full security model. Making it opt-in is the right call, and it should stay that way.

Open-weight models publish their trained weights so anyone can download and run them, though they often come with license limits that fully open-source software doesn't have. Opera has been experimenting with locally run models in its browsers for some time, and WebAI grows out of that work.