Opera WebAI: Local AI That Runs in Your Browser, Not the Cloud

Direct answer:

Direct answer: Opera has begun shipping local AI models directly inside its browsers through a new feature called WebAI, available first as an opt-in early preview in Opera One. Instead of sending your words to a distant data centre every time you use an AI feature, WebAI lets websites and browser extensions run a model on your own computer, with no AI subscription and no data leaving your

device for that step. It is a small, early release with a large idea behind it: a growing share of everyday AI work does not need the cloud at all.

What happened

On 27 October 2026, Opera announced that it had started shipping local AI models in its browsers, making AI that runs on the user’s own computer a central part of the next generation of its browsers. The first capabilities are available as an early preview in Opera One, the company’s flagship browser, with Opera GX and Opera Neon to follow. The release also includes HTML-in-Canvas, a separate feature that lets developers place live, interactive web pages inside two- and three-dimensional scenes, but WebAI is the headline for anyone thinking about how AI is delivered.

WebAI is described by Opera as a proposed web standard, backed by companies such as Google and Microsoft, that lets websites and browser extensions use AI hosted through the browser. Opera’s implementation is model-agnostic, meaning it is not tied to a single AI provider, and it is built around open models. At launch, users can draw on Google’s Gemma models and, on Macs, Apple’s on-device foundation model. Opera says more open models, tuned for different devices, are coming.

The announcement also matters because of who is saying it and how they frame it. Krystian Kolondra, Opera’s executive vice president for browsers and AI, put it plainly: much of what people want from AI does not need a data centre, and a modern laptop can already handle it without sending data anywhere. That is the pitch in one sentence, and it is worth taking seriously, because it points at a genuine shift in where AI runs, not just a new button in a browser.

This is not Opera’s first step into local models. The company says it was the first major browser to offer built-in access to local AI models, in an experimental developer build in 2024. That earlier build offered experimental support for around 150 local model variants from roughly 50 model families, including Meta’s Llama, Vicuna, Google’s Gemma and Mistral’s Mixtral, each typically needing 2 to 10 GB of local storage. Those models sat alongside Aria, Opera’s online browser AI service. What changes now is the audience: local models move from a developer experiment into an everyday browser, with a standard intended to let any website or extension use them.

Alongside WebAI, Opera shipped version 139 of its stable desktop browser, updating Chromium and adding the WebAI preview, the HTML-in-Canvas flag, a refreshed start-page personalisation widget and various polish fixes. The browser update itself is routine; the strategic signal is not. Opera is betting that the browser, rather than a separate app or a cloud dashboard, becomes the place where ordinary AI runs.

How local AI in the browser actually works

To understand why WebAI matters, it helps to compare it with the AI most people use today. In the usual cloud setup, an AI feature on a website sends your text to a company’s servers, the model produces an answer there, and the answer travels back to your screen. Every prompt is a round trip, and somebody pays for the computing on the other end, usually through a subscription or a per-request charge passed on to the business running the website.

Local AI turns that journey inside out. The model itself lives on your computer, downloaded once and run by the browser. When a website or extension asks for help, the browser hands the task to the model on your machine. The data involved in that step does not travel to an AI provider’s servers. Opera’s own example is an extension that sorts your tabs, flags clickbait or spots relevant job posts using a model on your laptop, with no account, no fee and no data sent to an AI service.

WebAI covers three kinds of capability at launch: generating text, classifying content, and understanding how pieces of information relate to each other. In plain terms, a website or extension can ask the browser to write something, sort something into categories, or work out connections between items. Opera has also released a collection of example extensions to help developers and users explore what is possible, and the preview includes a playground of on-device functions covering prompting, summarising, writing, rewriting, translating, proofreading, embedding and decision-making.

There are two important technical details that keep this honest. First, WebAI is a proposed standard, not a finished, universally adopted one. Opera says companies such as Google and Microsoft back it, but a proposed standard still needs other browsers and developers to adopt it before it becomes something every website can rely on. Second, “local” describes where the model runs, not a guarantee about everything a website does. A site using a local model could still send other data elsewhere for its own purposes, which is why Opera says users will always decide what runs on their machine and which websites may access local models. The privacy gain is real, but it is a gain in one specific step, not a blanket promise.

Hardware is the other practical constraint. Local models need a reasonably capable machine. At launch, WebAI targets 64-bit Windows computers and Macs running Apple silicon, and the models themselves take storage space and processing power. Older or low-spec machines will struggle, and the largest, most capable models still belong in data centres. That is why the sensible reading of this launch is not “the cloud is dead” but “the browser can now handle a useful slice of AI on its own.”

How WebAI compares with the alternatives

The most obvious comparison is with cloud AI services such as the chatbots and writing assistants most businesses already pay for. Cloud AI wins on raw capability: the biggest models, the longest reasoning, the broadest knowledge, and no need for powerful local hardware. It loses on cost predictability, on data leaving your premises, and on dependence on a provider’s uptime and pricing. If a provider raises prices or changes terms, every prompt gets more expensive.

A second comparison is with on-device AI already built into operating systems and phones. Apple, for example, has been pushing its own on-device foundation model on Macs and iPhones, and Microsoft has promoted on-device Copilot features on newer Windows machines. Opera’s move is different in an important way: it puts local AI into the browser layer, where websites and extensions can reach it, rather than locking it inside one operating system’s apps. If the standard spreads, a small business could build one browser-based AI feature and have it work wherever the standard is supported, instead of building separate versions for each platform.

A third comparison is with Opera’s own earlier position. In 2024, local models in Opera were a developer curiosity: interesting, fiddly, and limited to people willing to download multi-gigabyte models and experiment. The 2026 release packages that idea for ordinary users, with an opt-in preview, a choice of models, and example extensions. The gap between those two stages is the gap between a demo and a product, and it is the reason this story deserves attention even though the feature is still labelled a preview.

There is also a useful contrast with the infrastructure news dominating AI headlines this week. While other companies raise hundreds of millions to build data centres full of specialised chips, Opera is arguing that a meaningful amount of AI value can be created on hardware people already own. Both stories can be true at once: training giant models and serving millions of users at once still needs enormous infrastructure, while a single person’s drafting, summarising or sorting task does not. The industry is splitting into two layers, and WebAI sits squarely in the second one.

What it means for UK small businesses

For a UK small business, WebAI is not something to rush out and rebuild around tomorrow. It is something to understand now, because it changes the cost and privacy arithmetic of adding AI features to the tools you already use in a browser, and because the direction of travel is clear: more AI will run locally, and the businesses that prepare for that shift will spend less and worry less about data later.

Start with cost. The single most attractive claim for a small firm is that local AI removes the meter. Cloud AI pricing is usually a subscription per user, a charge per thousand tokens, or both, and those bills grow with usage in ways that are hard to forecast. A business that adds an AI summariser to its internal dashboard, for example, pays more every month as staff use it more. With WebAI, the model runs on the staff member’s own machine, so there is no per-request charge for that feature and no AI infrastructure to run. That does not make AI free: you still need decent laptops, time to test, and possibly a developer to build or adapt an extension. But it moves the cost from an open-ended monthly bill to a fixed, predictable investment in hardware you probably already buy. For a firm with ten or twenty staff, that difference is worth modelling properly before renewing a cloud AI subscription.

Next, consider data protection, which is where UK businesses have a concrete legal reason to pay attention. Under UK GDPR, a business is responsible for personal data it processes, including data it sends to a third-party AI provider. Every time an employee pastes customer details, supplier terms or staff information into a cloud AI tool, that business needs a lawful basis, a contract with the provider, and an answer to where the data goes and how long it is kept. Local AI does not erase those duties, but it shrinks the problem in a useful way: if the model runs on the employee’s own machine and the data in that step never leaves it, there is one fewer external processor involved. For tasks such as summarising internal notes, drafting replies to routine enquiries, or classifying support tickets, that is a genuine compliance advantage, provided the business checks what the website or extension itself does with data beyond the model step. The honest advice is to treat WebAI as a way to reduce exposure, not as a reason to stop doing data-protection checks altogether.

There is also a practical productivity angle. Because local AI does not depend on a network round trip to a provider, it can feel faster for short tasks and it keeps working when connectivity is poor, on a train, in a warehouse with patchy Wi-Fi, or on a site visit. A field engineer summarising notes, a shop owner drafting a product description, or a tradesperson sorting enquiries could do that work without waiting on a server or worrying about a connection dropping mid-task. The gains here are small per task but they compound across a team, and they arrive without any new subscription to manage.

For businesses that build or commission web tools, WebAI opens a new, cheaper route to adding AI features. Today, a small agency or freelancer building a browser extension or a web app with AI features must either pay a provider for every request or run their own servers. WebAI offers a third option: build the feature once, let it run on the user’s machine, and avoid both the infrastructure bill and the per-request meter. That could make AI features viable for niche tools that could never justify cloud costs, such as an extension that helps a specific trade sort job leads, or a dashboard widget that summarises a small firm’s own data. The catch is adoption: until other browsers support the standard, a feature built on WebAI will only reach Opera users, so early projects should be treated as experiments, not as replacements for a main product.

Customer trust is another angle worth weighing. Many customers are uneasy about their data being sent to AI companies, even when the business handling it has good intentions. Being able to say, truthfully, that a particular AI feature runs entirely on the customer’s own device is a clear, understandable privacy story, and it is much easier to explain than a data-processing agreement. A small business that offers a browser-based tool, a quote form, or a support widget could use that story as a differentiator, provided the claim is accurate and limited to the feature in question. Overclaiming here would backfire quickly, so the wording matters: “this summary is generated on your device” is a strong claim; “we never use AI” is a different and much riskier one.

Finally, there is a timing question. Early previews are, by definition, unfinished. Features change, models improve, and support can be withdrawn or reshaped. A small business should not bet a core workflow on an opt-in preview in a single browser. The sensible posture is to watch, test, and prepare: try the preview on a spare machine, see which everyday tasks local models handle well, note where they fall short, and keep cloud tools for the heavy work. When the standard matures and more browsers adopt it, the businesses that already know what local AI does well will be able to move quickly, while those starting from scratch will be learning the basics.

Honest limits and what to watch

The most important limit is capability. Open-weight models that run on a laptop are genuinely useful, but they are not the frontier models. They will be weaker at complex reasoning, at specialist knowledge, and at long, nuanced documents. A business that tries to replace a top-tier cloud model with a local one for contract analysis or detailed research will be disappointed, and should not be sold that story. The right test is task by task: drafting, summarising, sorting and short-form writing are strong candidates; deep analysis is not yet.

The second limit is adoption. WebAI is a proposed standard, and standards only matter when others adopt them. Opera says Google and Microsoft back the proposal, but backing a proposal and shipping it in your own browser are different things. Watch for announcements from other browser makers, and for developers actually building extensions on WebAI, rather than just applauding the idea. If, in six months, only Opera supports it, WebAI remains a niche feature for Opera users. If Chrome, Edge or Firefox move in the same direction, it becomes infrastructure.

The third limit is hardware reality. Not every machine in a small business will run local models well. Older laptops, machines with little storage, or shared computers in a shop or workshop may struggle, and the business will need a realistic view of its fleet before promising staff a local AI feature. Battery life is another factor on laptops: running a model locally uses processing power, and staff will notice if their machine slows down or drains faster. Testing on the actual machines staff use, not on a reviewer’s new laptop, is the only honest way to judge this.

Privacy also deserves a careful note. Local AI keeps the model’s input on the device, but the website or extension around it can still collect data for its own purposes, and users must be able to control which sites access local models. Businesses should check the permissions and privacy statements of any extension they adopt, and should not present WebAI as a guarantee that no data ever leaves a device. The gain is specific and valuable; it is not total.

Finally, there is the question of Opera itself. Opera is a smaller browser maker than Google, Apple or Microsoft, and its fortunes do not move the whole market on their own. A small business should follow the standard, not the brand: the idea of browser-hosted local AI is bigger than any single company, and the winners will be decided by which browsers and developers adopt it. Opera deserves credit for shipping first, but shipping first is not the same as winning.

What to watch, in short: whether other browsers announce WebAI support; whether developers release useful extensions beyond the examples; whether the range of available models grows quickly; and whether businesses report real cost savings from replacing cloud calls with local ones. Those four signals, over the next two quarters, will tell you whether this is a curiosity or a turning point.

Key facts

FAQs

What is Opera WebAI?

WebAI is a proposed web standard from Opera that lets websites and browser extensions use AI models hosted through the browser itself, running on the user’s own computer rather than on a remote server.

Which browsers and devices support WebAI at launch?

WebAI is available as an opt-in early preview in Opera One, with Opera GX and Opera Neon to follow. At launch it targets 64-bit Windows machines and Macs running Apple silicon.

Which AI models work with WebAI?

Opera’s implementation is model-agnostic and built around open models. At launch, users can use Google’s Gemma models and, on Macs, Apple’s on-device foundation model, with more models promised.

Does local AI mean my data stays private?

With local AI, the information the model handles stays on your own device rather than being sent to an AI provider’s servers. That reduces exposure, but it does not remove every privacy question: the website or extension using the model still matters, and users choose which sites may access local models.

Will WebAI replace cloud AI tools for small businesses?

No, not on its own. Local models suit everyday tasks such as drafting, summarising and classifying content. Larger, more demanding work will still run in the cloud for some time, so most businesses will use a mix of both.

What does WebAI cost a small business?

There is no AI subscription for the local models themselves, and developers do not need to run AI infrastructure or pay per request for WebAI features. The real costs are capable hardware, testing time, and any development work to build or adopt extensions.

Sources and further reading

• Opera’s own announcement of local AI in its browsers, including the WebAI preview and the quote from Krystian Kolondra, is the primary source for the launch details. See the Opera Newsroom release of 27 October 2026.
• TechCrunch’s 27 October 2026 report on Opera’s new web standard for locally deployed browser AI is the originating news report for this story.
• For context on where on-device AI fits, see our earlier coverage of Apple’s on-device AI features in its latest operating systems: https://weblifyai.co.uk/apple-ios-macos-27-2-siri-ai/
• For the wider agent-platform picture, see our coverage of a new AI agent platform launch: https://weblifyai.co.uk/strawberry-ai-agent-platform-launch/

CTA: Get a Free AI Consultation