Most AI features on the web work the same way. You type something, a request travels to a company's server, a model runs there, and the answer travels back. It works, but every request costs money, adds delay and sends your data somewhere else.
On October 27, 2026, Opera announced a different option called WebAI. It lets websites and browser extensions use an AI model that runs on the user's own computer, inside the browser. No cloud call is needed for the model itself.
If you build chatbots, automations or web tools, this is worth understanding now, even though it is still early. Here is what it is, what it can do, where it falls short, and where it could fit in real projects.
What Is Opera WebAI?
WebAI is a proposed web standard for local AI in the browser. Opera says it is backed by companies such as Google and Microsoft. Note the word proposed: this is not a finished standard that every browser supports. Right now, it is an early preview.
The idea is simple. The browser downloads and manages an open-weight model on the user's device. Websites and extensions then ask the browser to use that model through a set of APIs, instead of calling a paid cloud service.
Opera says its first version supports open-weight models such as Google's Gemma. On Macs, it also supports Apple's on-device foundation model.
What Can It Do Today?
According to Opera and the early coverage, WebAI can handle three kinds of work directly in the browser:
- Generating text
- Classifying content
- Understanding how pieces of information relate to each other (semantic embeddings)
Reports on the launch mention Prompt, Decision and Semantic Embeddings APIs. Opera's developer page also lists Summarizer, Writer, Rewriter, Translator and Proofreader APIs, with playgrounds that run on the device.
Opera also launched HTML-in-Canvas the same day. It is a separate feature, so we will leave it aside here.
How to Try It
The preview is available in Opera One first. Opera GX and Opera Neon are expected to follow.
Based on the launch coverage, the steps are:
- Update to the latest Opera One. For now, that means Windows (x64) or a Mac with Apple Silicon.
- Open Settings > Opera AI.
- Turn on Enable Web AI API.
- Let the browser download a local model.
Two details matter for trust. The feature is off by default, and no model is downloaded until the user turns it on. Users can also control which websites are allowed to use the local AI.
Cloud AI vs Local Browser AI
Neither option wins everywhere. Here is a fair comparison.
| Cloud AI API | Local AI in the browser | |
|---|---|---|
| Where the model runs | Provider's servers | User's own device |
| Cost per request | Pay per token or call | No per-request fee to a provider |
| Data path | Leaves the device | Stays on the device |
| Model strength | Strongest frontier models | Smaller open-weight models |
| Speed | Depends on network and load | Depends on the user's hardware |
| Reach | Works on any device | Only supported browsers and capable machines |
The last row is the big one. Today, local browser AI only reaches people who use a supported browser, on supported hardware, and who chose to switch it on. You cannot build a product that depends on it.
Where This Could Be Useful for Builders
WebAI is not a replacement for cloud models. It is more interesting as a first layer in front of them. A few ideas that fit its current strengths:
Small, private tasks. Proofreading a message, rewriting a paragraph or translating a short note. These do not need a top-tier model, and users may prefer that the text never leaves their device.
Pre-filtering before the expensive call. A local model can classify an incoming message (is this a support question, a sales lead, or spam?) before you decide whether a paid cloud model is needed at all. This is one practical way to cut spending. We cover more habits like this in our guide on how to control AI costs in chatbot, RAG and automation projects.
Search and matching on the device. Semantic embeddings in the browser could power quick "find similar" features over small amounts of data, such as a user's own notes or saved items, without building a server for it.
Graceful fallback. The safest design is progressive: use local AI when the browser supports it, and fall back to your cloud API when it does not. That way nobody is locked out.
The Limits You Should Know About
Be honest with yourself about the trade-offs before getting excited.
- It is a preview. APIs can change. Do not build critical features on it yet.
- Hardware matters. Running a model locally puts load on the user's computer. Older or weaker machines may feel slow, and Opera itself notes that performance depends on the device.
- Smaller models. Local models are generally less capable than the largest cloud models. Expect good results on simple tasks and weaker results on complex reasoning.
- Privacy is better, not perfect. Keeping processing on the device gives users more control, but it does not remove every privacy risk. Extensions and websites still need careful permission handling.
- One browser for now. A standard only becomes useful when more than one browser supports it. Watch whether Chrome, Edge and others follow.
What I Would Do Next
If you build web tools or automations, a sensible plan is small and low-risk:
- Try the preview yourself and note what runs well and what does not on your own machine.
- Pick one low-stakes feature, such as proofreading or classification, and prototype it with a cloud fallback.
- Keep your architecture flexible, so the model call sits behind one function you can swap.
- Check back in a few months to see whether other browsers adopt the standard.
Treat local browser AI as something to learn about now and adopt later. If you work with AI agents, remember that anything running with access to a user's files or tools needs the same care. Our checklist on AI coding agent security risks covers the mindset well.
Frequently Asked Questions
Is Opera WebAI an official web standard? Not yet. Opera describes it as a proposed standard. It is in early preview and only in Opera One for now.
Does WebAI send my data to the cloud? The model runs on your device, so the AI processing itself stays local. That gives you more control, but it does not remove every privacy risk.
Is it free to use? Opera says local AI lets developers add features without running AI servers or paying for every request. You still need a device capable of running the model.
Which devices are supported? At launch, Opera One on Windows (x64) and on Macs with Apple Silicon.
Will it replace cloud AI APIs? Unlikely. Cloud models are still stronger for hard tasks and work on every device. Local AI is best seen as a complement for simple, private or cost-sensitive tasks.
Final Thoughts
Local AI in the browser will not change how you build tomorrow. But it points in a clear direction: more small tasks handled on the user's device, and cloud models saved for the work that truly needs them. Learning how it behaves today gives you a head start when the standard matures.
Sources and further reading: Opera's announcement as reported by TechCrunch, CyberInsider and Thurrott. Details are based on the early preview and may change.