Run the model on the laptop, not on somebody else’s hardware. That is the wager behind WebAI, a browser specification Opera unveiled on Tuesday, and the trade-offs arrive with it: local inference taxes the machine, and Opera concedes that results will vary with the hardware underneath.
What buyers get in exchange is a stack that never leaves the device. Open-weight models can usually be downloaded at no cost, developers may customize them, and the computation stays put, so nothing goes to a vendor’s server and nothing is billed per request. Opera already sells itself on privacy, so the argument fits.
The company put it plainly. “Developers can add AI features without running AI infrastructure or paying for every request,” it said, adding that users dodge subscriptions and retain authority over the software on their machines.
Nothing in the spec is bolted to one vendor. Opera cites Google’s open-weight Gemma as one candidate and says WebAI also runs Apple’s on-device foundation model on Macs. The work builds on earlier Opera efforts to bring local models into its browsers.
An early preview is now live inside Opera One, with sign-up required. Opera GX, aimed at gamers and streamers, and the agent-driven Neon are next in line. On the capability side, a page could generate text, classify material and map how facts relate, all within the browser frame.
Privacy matters more as assistants reach further into personal data. Processing on the user’s own chip shrinks the exposure but does not erase it, because a browser still hosts the models and the pages that call them. Cheaper machines will also find heavier workloads out of reach.
