# Opera WebAI: Local AI in Your Browser, No API Key > Source: > Published: 2026-10-28 18:13:22+00:00 Opera shipped something developers have been waiting on browser vendors to build for years. **WebAI**, which landed in Opera One on October 27, lets websites and extensions call AI directly from the browser — no API key, no cloud backend, no per-request cost. The model runs on the user’s device: Google’s Gemma on Windows, Apple’s on-device model on Mac. It is still a preview, still Opera-only, and still covers roughly 3% of the browser market. But the standard it is proposing — and the W3C fight it is wading into — could reshape how the web handles AI. ## Eight APIs, Zero Infrastructure WebAI ships eight APIs. The three core ones are the Prompt API for text generation, the Decision API for content classification, and the Semantic Embeddings API for finding relationships between content. Rounding out the set: Summarizer, Writer, Rewriter, Translator, and Proofreader. The basic call looks like this: ``` js const session = await window.ai.createTextSession(); const result = await session.prompt("Summarize this: " + pageContent); ``` That is the whole integration. No auth headers, no rate limit logic, no billing dashboard to watch. For extension developers especially, this changes the economic calculation. Small teams have been shelving AI features for years because a successful launch could mean a surprise API bill. WebAI removes that ceiling. Tab organization, clickbait detection, semantic document search, job post filtering — all viable at zero marginal cost, if your user is on Opera. Opera has already shipped [six reference extensions](https://blogs.opera.com/news/2026/10/webai-local-ai-for-the-web-now-in-opera/) showing the pattern, including a quiz generator and an article relevance scorer. ## The Standards Fight Opera Is Entering WebAI is not just a product launch. It is Opera’s shot in a live W3C debate over who controls AI in the browser. When Chrome shipped its own Prompt API, it tied the implementation to Google’s Gemini Nano and required developers to accept Google’s Generative AI Prohibited Uses Policy to call a browser-level standard. Mozilla and Apple filed formal objections. The W3C Technical Architecture Group flagged what it called calcification risk — locking one vendor’s model into a web platform API would entrench that vendor permanently in web infrastructure. Opera’s answer is architectural: WebAI only supports open-weight models. No single company controls which model powers the API. No content policy gates developer access. [TechCrunch reports](https://techcrunch.com/2026/10/27/opera-introduces-a-new-web-standard-to-let-users-deploy-ai-locally-in-their-browser/) Google and Microsoft back the approach — though that claim comes from Opera’s own announcement, not a W3C filing. Apple and Mozilla still oppose. ## The Privacy Claim Needs a Caveat Opera’s messaging — and most of the press coverage — frames WebAI as a privacy win because data stays on your device. That is technically accurate and practically overstated. Local model inference means the AI computation happens on the user’s machine. It does not mean the website cannot read the AI’s output and transmit it elsewhere. It does not mean the extension code is trustworthy. Krystian Kolondra, Opera’s EVP of Browsers, said users can run AI without sending your data anywhere. That describes how the model call works — not what happens after it returns. Developers who pitch WebAI to users as a privacy feature should be precise about what that actually guarantees, which is [less than the marketing implies](https://www.techtimes.com/articles/329584/20261028/opera-webai-delivers-vendor-neutral-browser-ai-standard-without-chrome-content-policy.htm). ## Adoption Math: Where Things Actually Stand WebAI is available now, but the eligibility list is narrow. Supported hardware is 64-bit Windows (Intel or AMD) and Apple silicon Macs. Intel Macs, Linux, and Windows on ARM are not yet supported. Users must enable Web AI in Opera’s settings, then grant per-site permission. Models download at first use — 2 to 10 GB per variant. Realistically, you are building for a fraction of users on a 3%-market-share browser who have opted into an experimental feature. Any feature that depends on WebAI needs a graceful fallback. That is not a reason to ignore it — but it is the honest framing before you restructure your architecture around it. ## The Question That Actually Matters Google already ships local AI inference in Chrome — the Gemini Nano-powered Prompt API that drew all those objections. The real question is whether Google adopts WebAI’s model-neutral approach instead, or alongside it. If Chrome backs WebAI and drops the Gemini Nano lock-in, this standard instantly reaches 65% of desktop browsers. Google benefits from Gemini API revenue. Adopting a model-neutral browser standard is a business tradeoff, not just a technical one. What Google decides over the next 12 months will determine whether WebAI is a niche Opera feature or the foundation of how the web runs AI for the next decade. [Early developer analysis](https://dev.to/paullovelace/opera-now-ships-local-ai-in-its-browser-webai-means-extensions-without-the-api-bill-15) suggests the API design is solid enough to build on — the variable is adoption, not architecture.