THE ESSENTIALS IN 30 SECONDS
- WebAI exposes Prompt, Decision and Semantic Embeddings APIs that run on a local model, with no API key or cloud account.
- It is an opt-in early preview on 64-bit Intel/AMD Windows PCs and Apple silicon Macs; Opera GX and Neon follow later.
- Opera says Google and Microsoft back the proposal, but similar browser AI APIs have drawn objections over privacy and interoperability.
Opera has introduced WebAI, which it describes as a proposed web standard allowing websites and browser extensions to use an AI model that runs on the user's own computer. The feature arrived on October 27 as an early preview in Opera One, the company's main desktop browser.
How it works
Instead of calling a paid cloud API, a web page can ask the browser to run a task on a locally installed model. Opera's announcement lists Prompt, Decision and Semantic Embeddings APIs, meaning developers can generate text, classify information and find relationships between pieces of content. The implementation is not tied to one vendor: it runs open-weight models on Opera's inference engine, can use compatible models already on the device, and on Macs can also use Apple's on-device foundation model. TechCrunch cites Google's Gemma as one of the supported open models.
The feature is off by default. Users switch on "Enable Web AI API" in Opera's AI settings, which triggers a model download, and each site must be granted permission. Opera has published a developer playground and several demo extensions, including one that writes a quiz about the article being read.
What it changes
For developers, the pitch is that simple AI features no longer require running servers or paying per request. For users, Opera says content processed by the model stays on the machine and no AI subscription is needed. Opera also says Google and Microsoft support WebAI as a proposed standard.
Limits and open questions
The preview only runs on 64-bit Intel or AMD Windows PCs and Macs with Apple silicon; Intel Macs, Linux and Windows on Arm are not supported yet. Opera GX and Opera Neon are due to get it later, without a date. Performance depends on the hardware, and Opera has not detailed memory requirements.
Privacy is more nuanced than "nothing leaves the device": the model runs locally, but the website calling it still sees the page and could send results to its own servers. Comparable proposals, such as Chrome's Prompt API, have faced objections from other browser makers over privacy, battery use and the risk that sites behave differently depending on each browser's built-in model. Whether WebAI gains support beyond Opera will decide if it becomes a real standard.
