Opera is testing a new way for websites and browser extensions to use artificial intelligence without sending every request to a remote cloud service. The company's WebAI preview runs local models inside Opera One, the main desktop browser, and allows third-party developers to tap those models on the user's machine. The pitch is straightforward: routine AI tasks can be handled by the laptop or desktop already in front of the user, provided the hardware has enough spare capacity.
The feature is not a separate subscription product. Opera says WebAI does not require another AI plan and does not send the information handled by the models to a cloud service. Instead, the browser becomes the bridge between a website or extension and a local AI model. That arrangement could reduce latency, lower server costs for developers, and keep more sensitive data on the device.
Key facts at a glance
- Opera has introduced a WebAI preview that lets websites and browser extensions use local AI models inside Opera One.
- The preview is designed to avoid another AI subscription and to keep the information handled by the models from being sent to a cloud service.
- WebAI opens local AI to third-party developers rather than reserving it for Opera's own features.
- The user's laptop or desktop supplies the computing power, as long as it has enough capacity to spare.
- Opera describes WebAI as a proposed web standard that gives websites and extensions access to AI through the browser.
- Opera says the effort has backing from Google and Microsoft.
- The implementation is model-agnostic and centered on open models. Opera One can tap Google's Gemma models, while compatible Macs can also use Apple's on-device foundation model.
- Developers can generate or classify text without maintaining their own AI servers or paying for every request sent to an outside service.
- Opera's examples include an extension that sorts a messy tab collection and another that flags clickbait.
- WebAI is disabled by default. Nothing is downloaded until the user enables it, and each site needs permission before it can access the browser's local AI.
- Speed and capability depend on the computer. Opera has not disclosed minimum performance requirements beyond the supported platforms.
- Opera GX and Opera Neon are scheduled to receive WebAI later.
- Opera is pairing WebAI with a separate experiment called HTML-in-Canvas, which lets developers place functional webpages inside 2D and 3D scenes.
- The showcase includes a Gamer Room with live webpages on its screens and a Webpage Bulldozer that turns site elements into objects users can shove around.
A browser-level bridge to local models
WebAI is best understood as a browser-level bridge between web content and on-device AI. Rather than forcing every website to build its own connection to a cloud provider, the browser exposes local models through a standard interface. Opera calls WebAI a proposed web standard and says the effort has backing from Google and Microsoft. That backing matters because browser-based AI features only become broadly useful when developers can rely on consistent behavior across more than one browser and platform.
The implementation is model-agnostic and centered on open models. Opera One can use Google's Gemma models, and compatible Macs can also use Apple's on-device foundation model. The model-agnostic approach suggests that the browser is not trying to lock developers into a single AI vendor. Instead, it aims to provide a common layer that can route requests to whichever local model is available and appropriate for the task. For developers, the appeal is practical: they can generate or classify text without maintaining their own AI servers or paying for every request sent to an outside service.
That does not mean local AI is free or effortless. The computing power comes from the user's device, and the browser must manage downloads, storage, permissions, and performance. A model that runs comfortably on a high-end laptop may be impractical on an older machine. Opera has acknowledged that speed and capability depend on the computer, and it has not disclosed minimum performance requirements beyond the supported platforms. The company has also said that Opera GX and Opera Neon are scheduled to receive WebAI later, which indicates a staged rollout rather than an immediate switch for every user.
Why moving AI to the laptop matters
Cloud-based AI has become the default for many products because it centralizes computing power and makes advanced models available to devices that could not run them locally. But that model has trade-offs. Every request travels over the network, which can add latency and create bandwidth costs. Every request also creates a recurring inference bill for the developer or service provider. And every request may carry user data to a remote data center, raising privacy, compliance, and security questions.
Local AI changes that calculus. When a model runs on the user's machine, the information being processed can stay on the device. That can be attractive for tasks involving personal text, browsing behavior, or documents that a user would rather not upload. It can also reduce the number of network round trips, potentially making simple AI features feel faster and more responsive. For developers, local inference can shift some of the cost from servers to user hardware, which may make certain features economically viable without a subscription.
The trade-off is that local AI is constrained by the device. A laptop has finite processing power, memory, battery life, and thermal headroom. Running a model locally can compete with other applications, drain a battery, or cause a fan to spin up. The browser has to decide when local inference is appropriate, how to schedule it, and how to avoid degrading the overall experience. These are not just technical details; they affect whether users trust and tolerate AI features built into the browser.
Permission model and privacy controls
Opera has kept WebAI disabled by default. Nothing is downloaded until the user enables it, and each site needs permission before it can access the browser's local AI. That extra layer is important when websites, rather than only the browser vendor, can call the models running on the user's computer. A permission prompt gives the user a chance to decide whether a site's request is worth the compute, storage, and potential privacy implications.
The default-off design also addresses a common concern about AI features arriving uninvited. Users do not have to discover a model download in the background or wonder why a website suddenly has access to on-device intelligence. Instead, enabling WebAI is an explicit choice. Per-site permission adds another checkpoint, so a user can allow a trusted extension to sort tabs while denying access to an unfamiliar website. The model is that local AI is a capability the browser offers, not a background service that runs without consent.
Even so, permissions are only part of the story. Users will need clear information about which model is running, what data it processes, where the model comes from, and how much of the device's resources it uses. Opera has not yet detailed all of these operational questions. The early preview is likely to be a learning period for both the browser maker and developers, especially as they test how local AI behaves across different hardware configurations.
Practical examples: tabs and clickbait
Opera's examples are deliberately practical and a little nosy. One extension can sort a messy tab collection, while another can flag clickbait before it eats another five minutes of a user's life. These are small, everyday tasks that do not necessarily require a massive cloud model. A local model can classify text, recognize patterns, and make suggestions without sending the contents of every tab to a remote server.
The tab-sorting example shows how local AI could reduce browser clutter. A user with dozens of open tabs might benefit from an assistant that groups related pages, identifies duplicates, or suggests which tabs can be closed. The clickbait example shows how local classification could act as a lightweight warning system. Neither feature needs to be perfect to be useful, and both can be designed around user consent. The key is that the AI work happens close to the content and the user, rather than in a distant data center.
These examples also hint at the developer opportunity. If WebAI becomes a standard capability, extensions could offer AI-powered features without building a backend. A small team could ship a tab organizer, a reading assistant, or a content filter without paying for every inference request. That could lower the barrier to entry for AI features and encourage more experimentation. At the same time, it could lead to a flood of permission requests if developers treat local AI as a default add-on rather than a carefully considered feature.
Model flexibility and developer economics
The model-agnostic approach is one of the more consequential parts of the announcement. Opera One can tap Google's Gemma models, and compatible Macs can also use Apple's on-device foundation model. By supporting multiple models, the browser can adapt to different hardware and use cases. A lightweight model might handle text classification, while a larger model might be reserved for more demanding tasks. Developers could choose models based on quality, speed, size, or platform support.
For developers, the economic appeal is obvious. Maintaining an AI server requires infrastructure, monitoring, security, and a budget for inference. Paying for every request can make a free feature unsustainable. Local inference shifts those costs to the user's device and removes the per-request bill. That does not eliminate all costs, because developers still need to build, test, and maintain their extensions, but it changes the economics of small-scale AI features.
The model-agnostic strategy also creates challenges. Different devices will have different capabilities, and different models will produce different results. Developers may need to handle fallbacks, detect available models, and design interfaces that work across a range of performance levels. Browser vendors will need to provide clear APIs and documentation. A proposed standard with backing from Google and Microsoft could help, but adoption across the web takes time.
HTML-in-Canvas and the Webpage Bulldozer
Opera is pairing WebAI with a separate experiment called HTML-in-Canvas. The feature lets developers place functional webpages inside 2D and 3D scenes, where buttons and forms keep working as the page moves through virtual space. That is a notable shift from treating a webpage as a flat document. It suggests a future in which web content can be embedded in spatial interfaces without losing interactivity.
The showcase includes a Gamer Room with live webpages on its screens and a Webpage Bulldozer, which turns site elements into objects you can shove around. Both demos already work in Opera browsers without setup. That makes the bulldozer the part anyone can try while WebAI remains an opt-in early preview. The two experiments point in different directions: WebAI brings AI closer to the device, while HTML-in-Canvas brings web content into more dynamic visual environments.
The Webpage Bulldozer is more than a novelty. It demonstrates that browser-rendered content can be treated as physical material inside a scene. Buttons and forms still need to work, which means the browser has to preserve interactivity while moving and manipulating page elements. If that technology matures, it could affect gaming, virtual events, product demos, and other experiences that blend web content with 3D environments. For now, it is a demo that users can try without enabling WebAI.
What remains unclear
Several important questions remain unanswered. Opera has not disclosed minimum performance requirements beyond the supported platforms, so it is unclear which devices will offer a good experience. The company has also not detailed how model downloads are managed, how much storage they consume, how often they update, or how the browser will handle resource contention when multiple sites request local AI. These details will matter for users who expect AI features to be helpful rather than intrusive.
It is also unclear how quickly WebAI will move from preview to standard. Opera says the effort has backing from Google and Microsoft, but backing is not the same as broad implementation. Web standards require coordination, testing, and adoption across browser engines. Developers may wait for evidence that WebAI works consistently before committing to it. Users may wait for clear privacy controls and predictable performance before enabling it.
For now, WebAI remains an opt-in early preview inside Opera One. It is disabled by default, requires per-site permission, and depends on the user's hardware. Opera GX and Opera Neon are scheduled to receive it later. The HTML-in-Canvas demos, including the Webpage Bulldozer, are available to try without setup. The broader bet is that some AI work belongs on the device, close to the user and the data, rather than in a distant data center.
Source: Digital Trends News