Can Opera WebAI check your CV without uploading it?

What is Opera WebAI, and what changed on 27 October 2026?

Opera WebAI is a proposed web standard that lets websites and browser extensions use AI models running on your own computer, not on a remote server. Opera shipped it as an early preview in Opera One on 27 October 2026. It's model-agnostic, so it isn't tied to one AI system. The first models on offer are Google's Gemma and, on compatible Macs, Apple's on-device foundation model. For now the preview runs on 64-bit Windows and Apple silicon Macs. Opera GX and Opera Neon are due to get it later, and companies including Google and Microsoft have been named as supporting the standard. Opera isn't new to this, either. It first tested local language models in its developer channel back in 2024.

In practical terms, WebAI handles text generation, content classification and finding relationships between pieces of data. Opera also built a playground where you can try eight built-in browser APIs, including a Prompt API, a Summarizer API and a Writer API, plus translation and semantic embeddings. One of Opera's own example uses is an extension that highlights relevant job listings, which tells you the company already has job seekers in mind. The privacy defaults are sensible. WebAI is off until you turn it on, no model downloads until you do, and you grant each website access to the local AI separately. You decide which pages get to use it, and you can take that access away.

Why should someone polishing a CV care about a browser standard? Because a CV is one of the densest bundles of personal data you'll ever paste into a text box. It holds your full name, phone number, email, often your home address, ten years of employment history and sometimes your salary expectations. Lots of job seekers already paste all of that into a cloud chatbot without thinking twice. Others refuse to touch AI at all for exactly that reason. WebAI offers a middle path: the model sits on your machine, so the text it reads doesn't have to go anywhere. There's one honest caveat, which Opera itself makes. Processing data on the device doesn't guarantee how a particular website or extension handles it.

Is it safe to upload your resume to AI, and does WebAI fix the risk?

Uploading your resume to an AI tool isn't automatically dangerous, but it does put a copy of your personal data on someone else's servers, where their retention and training policies apply. WebAI changes that because the model runs locally. When a page uses the browser's built-in AI, the model itself has no reason to send your CV to a data centre. That's a real improvement over pasting everything into a cloud chat, especially if you're applying while still employed and your CV names your current employer, your team and the projects you're working on now. It also helps if you work in a regulated industry where even a job title and client list can count as sensitive information.

Before you get too relaxed, look at what's actually risky on the page. Usually it isn't the bullet points. It's the header: your full home address, date of birth, ID or passport numbers, a photo, and sometimes a personal phone number tied to every account you own. Most of that shouldn't be on a modern English-language CV anyway. City and country, one email and one phone number are plenty. Here's our blunt view: the most sensitive thing on most CVs is a line that never should have been there, and deleting it does more for your privacy than any choice of tool. If you're weighing up the big cloud assistants, we've also broken down which fields matter when you hand Meta Muse your CV.

The caveat that matters most with WebAI is that the model being local doesn't mean the page is. A website or extension that calls the browser's AI can still send what you typed to its own server for logging, analytics or some other feature. So treat WebAI permissions the way you'd treat microphone or location permissions. Only grant them to tools you trust, and check the list now and then. One quick way to test a local-only tool is to load it, turn off your Wi-Fi and see whether it still rewrites your text. If it works offline, the processing is happening on your machine. It doesn't prove nothing gets sent later, but it's a far better sign than any marketing claim.

What can a local in-browser model actually do for your CV?

A local in-browser model is good at sentence-level CV work: proofreading, rewording weak bullets, tightening a professional summary, translating sections and giving you rough first-pass feedback, all without your text leaving the device. Those jobs line up neatly with the Prompt, Writer and Summarizer APIs WebAI ships with. You don't need a frontier-scale model to spot that "responsible for" is a weak opener, or that three bullets in a row start with "Managed". A compact model like Gemma handles that kind of tidy-up well. It's also quick, because there's no round trip to a server, and you can repeat it as often as you like without hitting usage caps or paying for each request.

Concrete prompts work best. Paste one bullet at a time, something like "Responsible for managing the company's social media accounts", and ask: "Rewrite this as one CV bullet starting with a strong past-tense verb. Don't add numbers; put [X] where a metric should go." You'll get something like "Grew and managed four social channels, increasing engagement by [X]%", and you fill in the real figure yourself. Then try "List any spelling, grammar or tense inconsistencies in this section" for proofreading, or "Cut this summary to three sentences without losing the job title" for your opening lines. For bilingual applicants, including Hebrew speakers applying in English, local translation gives you a private first draft. Have a fluent reader check it before you send anything.

Expect the quality to drop as the task gets bigger. Small local models handle a single bullet or paragraph far better than a whole two-page document. Ask one to restructure your entire CV and it'll lose track of dates, merge roles or quietly invent achievements. The rule we'd insist on is simple: never keep a number the model came up with. If it writes "cut costs by 30%" and you didn't give it 30%, delete it. A recruiter can ask about any figure in an interview, and a made-up one will sink you quickly. If you'd rather run a bigger open model offline on a Mac, we've covered the hardware a local CV review really needs.

Where does a local AI CV review fall short?

A local AI CV review can't do three things that matter most at the screening stage: it can't score how an applicant tracking system will parse your actual file, it can't compare you with other candidates for the same role, and it can't reliably tailor your CV to a specific job ad. These aren't bugs that a future update will fix. They come from what the setup is. A small model reading text you've pasted into a browser box has no access to the file the employer will receive, no data on what strong CVs in your field look like, and only a limited ability to keep a long job description and a long CV in view at the same time.

The ATS gap is the one that catches people out. Parsing happens on the file itself, the PDF or Word document you upload to a careers portal. Two-column templates, text boxes, icons, tables and contact details tucked into the page header can all scramble the extracted text or drop it entirely. When you copy your CV into a WebAI prompt, you're pasting text that has already been cleanly extracted, so the model never sees the mess a parser might make. It can tell you a bullet reads well. It can't tell you the bullet ended up in the wrong field, or that your job titles vanished because they sat inside a graphic. That needs a tool that actually reads your layout.

Market benchmarking is the other blind spot. Whether your CV is "good" depends on the role, the seniority and the competition. A mid-level product manager's CV gets judged differently from a graduate analyst's. A local model has no reference set to compare you against, so its feedback falls back on generic advice. We'd go further: a small model's praise means almost nothing. These models tend to be agreeable, so "this looks strong!" is closer to a polite reflex than an assessment. They're great for wording, and you shouldn't count on them for judgement. Treat local feedback as a spell-checker with opinions, and get a proper verdict from something built to grade CVs.

What's the best privacy-first workflow for checking your CV?

The best privacy-first workflow is to do your drafting and polishing locally, strip the sensitive details, then run one dedicated CV analysis on the finished file for the ATS and job-match check. That way the messy, repetitive work stays on your device, and you share only one cleaned-up version once you're ready to apply. Most people get this backwards. They paste raw drafts into cloud chatbots twenty times over, then never check the final file at all. Turned the right way round, the workflow gives you more privacy and better results, because the check that matters most happens on the exact document a recruiter's system will receive.

Here's the sequence. First, turn on WebAI in Opera One and grant it only to the tool you're using. Work through your CV one section at a time with the proofreading and bullet prompts above, filling in real numbers yourself. Second, scrub the header. Remove your home address, date of birth, ID numbers and photo, and keep your city, one email and one phone number. Third, export to PDF or Word, whichever format the job ad asks for. Fourth, upload that file once to a free CV analysis. You'll get a score out of 100, five category scores including ATS compatibility, your strengths and a visual layout analysis that shows the formatting problems a local model can't see.

To go further, paste the job description into the analysis. Paste the text itself rather than a LinkedIn or Indeed link, because those sites block fetching. You'll see which skills the ad asks for that your CV doesn't show, and you'll get rewritten, ATS-friendly versions in six templates that use the job's keywords. If you need the full fix list and the "why you might be rejected" breakdown, the paid Full Analysis costs $3.99. Then go back to your local model to rework the weak sections, with no further uploads needed. If your old CV is beyond saving, start fresh with the AI CV builder, and if you want heavier cloud help for rewrites, here's how the Claude models compare on CV writing.

Frequently asked questions