Paste Guard features

Every feature in version 1.2.0, in one place. Everything runs on your device. Paste Guard makes no network requests and has no account or analytics.

Protecting a paste

  • Checks every paste on the supported AI chat sites. Clean text is pasted exactly as normal.
  • Warning dialog listing each unique item found, with a short preview (only the first two characters) and a before and after example.
  • Three choices: Paste masked (default, Enter), Paste anyway, Cancel (Esc). It never blocks you from the final choice.
  • Numbered placeholders such as [EMAIL_1], [CARD_1], [API_KEY_1]. The same value gets the same number, so the AI can still follow your text.
  • Ignore this item on any row. Only a salted one-way hash is kept, never the value.
  • Modes: Ask, Auto-mask or Off, plus a master on and off switch.
  • Safe by design: if anything fails internally, your original text is pasted and later pastes keep working. Large pastes (1 MB) are scanned in about a tenth of a second.
  • Works in textareas, inputs and rich text editors, keeping line breaks.

What it detects

  • Email addresses, phone numbers (anything starting with +, Indian mobiles), card numbers (Luhn checked).
  • Indian IDs: Aadhaar (Verhoeff checked) and PAN.
  • Regional and banking IDs: UPI IDs, IFSC codes, GSTIN (checksum verified), labelled bank account numbers, IBANs (mod-97 verified), UK National Insurance numbers.
  • US Social Security numbers.
  • API keys and secrets from about 75 providers, JSON Web Tokens, private keys, passwords in connection strings and labelled secrets.
  • Your own words, from the personal dictionary.
  • Every detector has its own switch in the popup. See What it detects for the full list and defaults.

Restore original values in replies

  • The AI only ever receives placeholders. When its reply uses them, your browser shows your original values again, on your device only.
  • On by default, with a switch in the popup.
  • The same value keeps the same placeholder across pastes in a session, per site.
  • The list that pairs placeholders with values is kept in memory only. It is never written to disk, expires after 4 hours, holds at most 500 entries per site, and is cleared when you close the browser.
  • Forget stored values now in the popup, and a count of restorable values.
  • Your message box always keeps the placeholders, so an original value cannot be sent by accident.
  • A small pill shows "Restored N values". Press Alt+R (Option+R on a Mac) or the pill button to switch between original values and placeholders. You can dismiss the pill.
  • Handles replies that stream in small pieces, placeholders split across elements, and placeholders inside bold text or code.
  • To find placeholders, the extension reads the page text on the supported sites. It sends nothing anywhere.
  • You can try the whole flow in your browser, with nothing installed, on the Try Restore and Mask this page.

Mask this and the personal dictionary

  • Select words in the message box and press Alt+Shift+M (Option+Shift+M on a Mac), or use the Mask this chip that appears next to the selection.
  • The words become a [CUSTOM_1] style placeholder, are restored in replies, and are added to your personal dictionary.
  • Matching in future pastes is whole word and ignores case, so "Zorbex" is found but "Zorbexville" is not.
  • Free limit of 10 words, then the extension says "Pro unlocks unlimited words". Pro is not available yet.
  • The words are stored on your device only. List, remove or clear them in the popup.

The popup

  • Protection switch, mode selector and a counter of pastes protected.
  • A switch per detector, grouped as Secrets and keys, Identity, Contact, Payments and Regional.
  • Restore switch, restorable value count and the Forget button.
  • Personal dictionary list with remove and clear all.
  • Ignored items count with a clear button.
  • A locked Pro section for custom regular expressions (not available yet).
  • Light and dark mode, keyboard friendly, reduced motion respected.
  • Version number in the footer, so you can tell which build you are running.

Welcome page

On first install a page opens with a practice box that runs the real detectors and the real dialog on made-up data, then shows a made-up reply with values restored so you can see the idea. It makes no network requests. The same walk-through is on the website as the interactive tour, and there is a short video tutorial.

Privacy and safety

  • No network requests, no analytics, no remote code, no account.
  • Permissions: storage plus the eight supported sites. Nothing on any other site.
  • Stored on disk: your settings, the paste counter, ignore hashes and dictionary words. Original sensitive values are never written to disk, page attributes, URLs or the console.
  • See Privacy and how to verify it yourself.

Supported sites

ChatGPT (chatgpt.com), Claude (claude.ai), Gemini (gemini.google.com), Perplexity (perplexity.ai), Copilot (copilot.microsoft.com), Grok (grok.com), DeepSeek (chat.deepseek.com) and Mistral (chat.mistral.ai).

Not available yet

  • Pro: unlimited dictionary words and custom regular expressions.
  • Detection is pattern based and can miss unusual formats. It does not read images or files, and it does not take back anything already sent.

Try the scanner