Opera announced on Tuesday (October 27) the launch of a pilot implementation of WebAI — a proposed standard that would allow web pages and extensions to access artificial intelligence models running locally on the user's device. The feature is available as an experimental preview in Opera One, with plans to roll out to Opera GX and Opera Neon in the coming weeks.
WebAI is not a product exclusive to Opera. It is a community-driven standard proposal, backed by companies including Google and Microsoft, aiming to establish a unified API that would let any website or extension use AI models hosted directly by the user's browser. Opera's implementation is, however, the first public, large-scale deployment of the technology.
How it works
The premise is straightforward: rather than sending queries to remote servers (as happens with ChatGPT, Claude, or Gemini), the AI runs directly on the user's computer. This means that personally processed data never leaves the device, there's no need to create accounts or provide API keys, and the user maintains full control over which models are running.
According to Opera, the implementation is "model-agnostic" — meaning it's not tied to a single vendor. At launch, users can already access models from Google (such as Gemma, released two years ago) and, on Macs with Apple Silicon, Apple's on-device foundation model. The company says the platform was built to support a wide range of open-weight AI systems.
The technical architecture relies on existing technologies: the implementation leverages WebGPU, the web's graphics and compute API, to accelerate model inference in the browser without burdening the main process. This allows models that previously only ran in data centers to be executed on consumer hardware — provided the device has sufficient GPU resources.
Opera's strategy
Opera wasn't the first to explore local AI. In April 2024, the company had already launched an experimental build for developers that allowed loading language models locally in the browser. At the time, the response was cautious — the required hardware was limited, models were small, and practical use was restricted to enthusiasts.
Three years later, the landscape has shifted. Models like Google's Gemma and Apple's on-device foundation model have demonstrated that the quality of locally running models approaches cloud versions for everyday tasks: text generation, content classification, understanding relationships between information. At the same time, the drop in hardware costs — especially with affordable GPUs in mid-range laptops — has expanded the potential user base.
The difference now is that Opera is packaging the technology directly into its consumer browser, not as a developer experimental feature, but as a core functionality. The company describes itself as a "browser and AI agent company" — and the strategy of positioning local AI as one of its product pillars is clear: create a barrier to entry against the big cloud players.
Krystian Kolondra, Opera's EVP of Browsers and AI, said in a statement: "We first shipped local models to developers in 2024. Now we're bringing them to everyone. More models and tools are coming, and you'll always decide what runs on your computer."
Advantages and limitations
The privacy promise is the biggest differentiator: since the AI runs locally, there's no sending queries to third-party servers. This eliminates the risk of leaking personal data to AI companies, a topic increasingly relevant amid recurring scandals involving data misuse.
For developers, the proposal is equally attractive. According to Opera, adding AI features to a website would no longer require maintaining inference infrastructure or paying per-request costs — the user loads the model, the site simply invokes the API.
But there are trade-offs. Opera acknowledges that running AI models locally imposes significant demand on the user's hardware. Larger models require more GPU memory and processing power, which can affect overall computer performance. Model performance also varies based on the user's hardware capabilities — a basic laptop may run Gemma at 2 billion parameters, but not larger models.
Additionally, WebAI is still a proposed standard, not a W3C-approved specification. This means the API may change before becoming definitive, and that adoption by other browsers depends on community will. So far, Google and Microsoft are the main supporters.
What this means for the future
WebAI's arrival in production marks a turning point: local AI moves from a lab experiment to a mainstream consumer experience. In a market where most AI interactions still flow through clouds controlled by tech giants, Opera is betting that decentralizing inference could be a competitive differentiator.
If the standard gains adoption by other browsers — and if consumer hardware keeps pace with growing demand — we could be looking at a new paradigm: browsers that don't just render pages, but that are active AI execution platforms, with privacy and user control at the center of the design.
The question is whether the big cloud players will accept competing on their own turf — or whether they'll try to keep AI centralized, using their scale advantage to make the cloud cheaper and faster than any local alternative.
Sources: Opera Newsroom, TechCrunch, 9to5Mac
✓ Independent sources cross-checked and verified before publishing
