Opera Ships WebAI to Run AI Models Locally in the Browser

Opera has shipped WebAI, a proposed web standard that lets websites and extensions run AI models on the user's own machine instead of a cloud provider's servers. The feature is now in early preview across Opera One, Opera GX and Opera Neon, and it hands AI builders something they rarely get for free: an inference runtime that already ships inside the browser.

What Opera shipped

WebAI exposes three JavaScript APIs. The Prompt API generates text from a task description. The Decision API answers yes/no, multiple-choice and scored questions about a piece of text in a single pass, with a confidence threshold. The Semantic Embeddings API turns content into vectors so code can measure how closely two passages relate.

There is no separate app to install, no local inference server to configure and no API key to manage. Local inference runs inside the browser on Opera's own engine, using open-weight models the company provides, such as Google's Gemma family, or compatible models already sitting on the device. On Apple silicon Macs it can also reach Apple's on-device foundation model.

"Developers can add AI features without running AI infrastructure or paying for every request," Opera wrote in its announcement. "Users don't need AI subscriptions, the information the local AI handles stays on their own device."

Why builders should care

For anyone shipping automation, the interesting part is not the chat window. It is the cost model. Cloud inference bills per token, which makes always-on features hard to justify: a classifier that runs on every page view, an embedder that indexes a user's reading history, or an agent loop that reacts to interface events. WebAI moves all three into the client, where the marginal cost of a request is zero and the data never leaves the machine.

Opera demonstrated the idea with six experimental extensions built by its own team. One, Project Lens, scores every paragraph of an article against whatever project the user is working on and colors it by relevance. Cheap to describe, awkward to run at cloud prices.

The proposed standard is also backed by companies including Google and Microsoft, which matters more than Opera's own preview, since a shared browser API is what turns local inference from a vendor trick into a deployment target.

The limits are real

WebAI is opt-in. Nothing downloads until a user enables it in Settings, so reach inside the browser is a user decision rather than a default. Coverage at launch is narrow: Windows PCs with 64-bit Intel or AMD chips, plus Apple silicon Macs. Linux, Windows on Arm and Intel-based Macs are excluded. Latency depends entirely on local hardware.

The builder takeaway

Treat WebAI as a zero-cost tier in front of your stack: when the Prompt, Decision or Embeddings API is present, run the cheap classification and embedding work locally, and fall back to your server when it is not. The browser just became one more inference target to route to.