Sunda Privacy Guard is a Chrome extension for recruiters who put candidate CVs into ChatGPT or Claude every day. Before a CV leaves your computer, it swaps the name, phone number, email, address and ID number for placeholders like [PERSON_1] — on your device, with nothing sent anywhere. You paste the clean copy, screen as usual, and keep the only key that turns placeholders back into people.
Your details never leave your device · No account · No tracking
Hand Sunda a CV with its real header block. It catches each identifier and hands back a clean copy with typed placeholders instead — the CV still reads like a CV, the screening still works, and the candidate's name, number and address stay on your machine.
The candidate's CV: Jordan Reyes. Senior Accountant, 8 years, ACCA. +44 7700 900123, jordan.reyes@example.com. 14 Maple Street, Manchester M1 4AB. Finance lead at Northwind Retail.
What the AI sees: [PERSON_1]. Senior Accountant, 8 years, ACCA. [PHONE_1], [EMAIL_1]. [ADDRESS_1]. Finance lead at Northwind Retail.
The phone and the email? Caught by the always-on rules, instantly, in every flow —
Israeli ID numbers are even checksum-verified before they're hidden. The name and the
address? That's Sunda's optional on-device AI model pitching in when you protect a
file in the popup — this peek shows it switched on; the rules catch the structured
details either way. The employer stays, on purpose: a CV with no companies on it is no
use to you. And if you choose the optional key, every placeholder can be turned back
into the original. By you, not by the AI.
A fifteen-second animation, a minute-and-a-half walkthrough, and three thirty-second spots. Nothing downloads until you press play, and every frame is served from this site — no third-party player, no embed watching you watch.
[BRAND_FILM] 0:15
Fifteen seconds, no words. A candidate's header block arrives, each identifier turns into a typed placeholder, and Sunda shows up to take the credit.
[WALKTHROUGH] 1:37
The whole tour: what leaks when a CV goes into a chat window, what Sunda swaps for what, where the decryption token lives, and how guarding uploads on a site you choose actually behaves.
[SPOT_1] 0:30
A recruiter, a CV and an hour to go. The file gets cleaned on its way into the AI and the shortlist still gets written.
[SPOT_2] 0:30
Shot from the file's side of the desk: a name, a mobile number and a street address going amber one at a time, then leaving as placeholders.
[SPOT_3] 0:30
A scripted advert. The recruiter is played by an actor — not a customer, and not an account of anyone's real week. The steps it shows are the same three steps on this page.
Nothing in the films is new: every claim they make is written out in text on this page and in what Sunda can't do. None of them carry subtitles yet.
No account, no cloud, no peeking. Sunda checks the CV right there in your browser — then hands you the clean copy with a grin.
Pick the file in the popup — Word, text, or a PDF you can select text in — or attach a Word or text file on a site you've asked Sunda to guard. It scans on your device for names, phone numbers, emails, addresses, ID and card numbers. Nothing is sent anywhere to be checked.
Each detail becomes a typed placeholder like [PERSON_1]. The same value always gets the same placeholder, so the AI can still say “Candidate [PERSON_1] has eight years in finance” — and you know exactly who that is.
Turn on reversible encryption and the protected file comes with a decryption token. Paste it on the Restore page to rebuild the original — on your device. Lose it, and not even Sunda can bring the original back.
If you've ever pasted a CV into a chat and thought “wait — was the phone number in that?”, this is for you. Protect the file first, then prompt as usual. (Phones, emails, IDs and links are handled by the always-on rules in every flow; names and addresses are the job of the optional on-device model when you protect a file in the popup.)
A five-bullet summary and three questions to ask — without the candidate's name and number sitting in a chat log for the next thirty days.
Summarise [PERSON_1]'s CV in 5 bullets and suggest 3 screening questions.
Paste the redacted CV and the JD; ask for fit, gaps and a score. The model needs the experience, not the identity.
Score this CV against the job description below (1–10) and list the gaps.
Turn interview notes into a kind, specific email — with [PERSON_1] as the stand-in you swap back before you hit send.
Draft a warm, specific rejection to [PERSON_1] based on the notes below.
Messy notes in, structured scorecard out: criteria, evidence, rating. The candidate's identifiers never leave your machine.
Convert these notes on [PERSON_1] into a scorecard: criterion, evidence, 1–5.
Build a comparison table of three redacted CVs and let the manager react to the experience first, the names second.
Compare [PERSON_1], [PERSON_2] and [PERSON_3] for this role in a table.
Add drive.google.com or your ATS to Protected sites and Word and text CVs get a rules check the moment you attach them — phones, emails, IDs and links swapped before the site sees a thing (names need the popup flow).
Protected sites → drive.google.com · chatgpt.com · claude.ai
A recruiter needs to know exactly what is detected, what isn't, and what risk remains. Here it is, before you install anything.
linkedin.com/in/name).Sunda reduces exposure. It does not make you compliant with anything. No GDPR badge, no Amendment 13 seal — using Sunda does not satisfy a legal obligation. It removes identifiers from files before they leave your device; the rest of your obligations are still yours.
Six screenshots, zero technical skills required. Every one comes straight from the real extension that's headed to the Chrome Web Store. Here's what living with Sunda looks like, from your very first click:
Pin Sunda next to your address bar and pop it open. Everything lives in one panel: protect a file in a click, pick the sites where uploads get guarded, flip on the undo key. The guard is up from the moment Sunda moves in — and if your computer supports WebGPU, Sunda grows a little on-device AI model that also spots names, addresses and dates in files you protect by hand. One tap — or the first time you protect a file — starts that one-time download, with a progress bar right in the popup. The built-in rules — emails, phone numbers, IDs, cards, links — guard you either way.
Pick a Word, text or PDF file and Sunda hands back a clean copy — plus a small report of what it tucked away, like “Replaced 3 items — 1 email, 1 phone number, 1 Israeli ID,” and an honest note about which engine did the checking. Straight to your Downloads, never to the internet. (The first protect after a browser start can take a little while — Sunda's AI model is still waking up, and it waits so your file gets the full treatment.)
Turn on reversible encryption and every protected file comes with its own decryption token — the key that can bring the original back. Copy it somewhere safe and treat it like a spare house key: it's the only one there is, and Sunda never keeps a copy.
Ready to call the candidate? Open Restore, load the protected file (the .spg or protected Word file you downloaded), paste your token, and the original is rebuilt right on your device. No cloud, no waiting, no drama.
Flip the switch, and Word and text files get a quick rules check the moment you attach them on a site from your list — chatgpt.com, claude.ai, Drive, your ATS — with emails, numbers, IDs and links swapped out before the site sees a thing. And if a file Sunda guards can't be checked, it blocks the upload instead of letting it slip — better a pause than a leak. One honest exception: once your free trial ends without a Pro license, guarded uploads aren't protected at all — the file goes through untouched, with a notice on screen telling you so.
Uploads get guarded only on the sites you add — nowhere else, nothing sneaky. Type a site, click Add, and Chrome itself asks you to allow it, one site at a time. Change your mind? One click on Remove and Sunda tips its sunglasses and steps away.
Under the friendly exterior, Sunda has one job and does it stubbornly well: the candidate's details in your files never reach the AI. Here's the tour, jargon-free.
Pick a Word, text or PDF file in the popup and get a clean copy back — identifiers out, placeholders in. Word keeps its layout; a PDF is rebuilt from its text, so you get the words and page breaks without the original styling. On computers with WebGPU, Sunda's on-device AI model joins in to catch names, addresses and dates too.
Add chatgpt.com, claude.ai, Drive or your ATS to Protected sites, and Word and text files get a quick rules check right as you upload them there. Only on your list — with an empty list, Sunda touches no webpage at all.
Choose protection with a key and you get a protected file plus a decryption token that turns every placeholder back into the real thing. For you — not for the AI.
Teudat Zehut numbers are checksum-verified to be real IDs, and Israeli phone formats are spotted too — alongside emails, cards and bank account numbers.
No guessing games with candidate data. On your Protected sites, if a Word or text file Sunda guards can't be checked, it simply doesn't get uploaded. Better a pause than a leak. One exception, and the extension says so on screen: once your trial ends without a license, protection stops rather than your uploads breaking — the file goes through as-is, with a clear notice.
No account, no telemetry, nobody peeking. The spotting and the swapping happen inside your own browser. The only things Sunda ever fetches are its own AI model (once) and, if you buy a license, a quick key check — never your files, never their contents.
[PERSON_1]
Emails
[EMAIL_1]
Phone numbers
[PHONE_1]
Israeli IDs
[ISRAELI_ID_1]
Credit cards
[CREDIT_CARD_1]
Bank accounts
[IBAN_1]
Addresses
[ADDRESS_1]
Dates
[DATE_1]
Links
[URL_1]
The same email always becomes [EMAIL_1], and a second one becomes [EMAIL_2] — so the CV still reads naturally, to you and to the AI. Israeli IDs are checksum-verified to be real ones, and Israeli phone formats are understood out of the box. Names, addresses and written-out dates are the specialty of Sunda's optional on-device AI model, which pitches in on files you protect by hand — everything else is caught by the always-on rules, every time.
It would be pretty silly to build a tool that protects candidates' private details… by peeking at candidates' private details. So these aren't pinky promises in a policy — they're simply how Sunda is built.
Your files, and everything found inside them, stay on your device. Sunda has no server of its own — the only things it ever talks to the internet for are fetching its own AI model and checking a license key, and neither request ever carries your files or their contents.
No analytics, no telemetry, no usage stats — in the extension and on this website. Sunda never phones home about you; it wouldn't even know what to say. Even we can't peek, and that's exactly how we like it.
The secret key that can undo the protection exists in exactly one place: the token in your hand. It's never saved to storage and never sent anywhere — lose it, and not even Sunda can bring the original back.
Upload guarding happens only on the sites you add, and Chrome asks your permission for each one. Keep the list empty and Sunda runs on none of your pages — not quietly on some of them.
☀️ Live on the Chrome Web Store · v0.1.1
Install it, pin it next to the address bar, and the next CV you screen goes into ChatGPT without the candidate's name, number, email or address attached. Everything on this page is the shipping extension — the screenshots came straight from it.
No waitlist and no email capture — collecting your email would be a funny look for a privacy tool. Install it and it just works. ☀️