Opera WebAI runs Gemma locally in the browser, no API key
An opt-in Opera One preview lets any permitted site run Gemma or Apple's on-device model at zero cloud cost — plus fresh agent-skill repos to grab.
Copy markdown
Local models, no API key, no server
Opera's WebAI — an opt-in early preview in Opera One — lets any site you permit run open-weight models on your own machine: text generation, content classification, and relevance analysis, with no API key and no per-request bill. Nothing downloads until you turn it on.
Gemma now, Apple's on-device model on Macs
At launch it runs Google's Gemma on 64-bit Windows (Intel/AMD) and Apple-silicon Macs, and taps Apple's built-in foundation model on Macs. Opera says it's model-agnostic across open weights; Opera GX and Neon are 'coming soon.'
What you can ship this weekend
Site-scoped local inference means browser extensions and web apps that summarize pages, organize tabs, flag clickbait, or spot job posts — all client-side, at zero marginal cost. Opera is shipping example extensions you can fork.
The asterisk: a standard of one, for now
WebAI is pitched as a proposed web standard — reportedly backed by Google and Microsoft, though only Opera ships it today. It's off by default, and Opera warns the API surface will change. Prototype with it; don't put production on it yet.
Fresh skills to drop into your agent
New on GitHub: plannotator (MIT) adds browser-based review of agent plans and diffs for Claude Code, Codex, and Gemini CLI; claude-seo (MIT, ~22K stars) runs a full SEO audit in minutes via 26 sub-skills and 19 agents; ai-design-skills ships a landing-page-design skill for Claude Code, Cursor, Codex, and Windsurf.