AI data-loss prevention — detects and masks PII before it leaves your browser.
Users
3
store count
Rating
0.0 / 5
unchanged in 30 days
Ratings
0
total ratings
Store rank
#294K
#241K among extensions
Version
1.0.6
updated 17 Sep 2026
Users
monthly, 1 months
Latest, 15 Oct 2026:3 usersMove the mouse over the chart for any month
Ratings count
monthly, 1 months
Latest, 15 Oct 2026:0 ratingsMove the mouse over the chart for any month
The Chrome Web Store rounds user counts down to the leading digit, so the user line moves in steps. The ratings count is exact and is the better signal of real adoption.
About PiiI
PiiI catches personal information before it leaves your browser and lands in a
third-party AI service.
Paste a customer email into ChatGPT, attach a spreadsheet to Claude, or ask
Gemini to summarise a contract — it inspects the text on the way out, shows you
what it found, and offers to swap each item for a stand-in before you hit send.
WHAT IT DETECTS
• Email addresses
• Phone numbers
• Names and postal addresses (on-device ML model)
• Social security numbers and tax IDs
• Credit card numbers
• Passwords, API keys and tokens
• Account, passport, licence and ID numbers
• URLs and dates
WHERE IT RUNS
ChatGPT, Claude, Gemini, Copilot and Perplexity. The complete list of supported
sites — with the exact URL patterns each one matches — is here:
https://github.com/JaySmith502/PiiI-public/blob/main/SITES.md
HOW IT WORKS
1. Install PiiI and pin it to your toolbar.
2. Open a supported chat site and write your message as you normally would.
3. Press send. If it finds anything sensitive it opens a review panel listing
each detected item next to a suggested replacement.
4. Accept the suggestions, edit any of them, or send your original text
unchanged.
Pattern-based categories (email, phone, credit card, SSN, API key) work the
moment you install. Name and address detection becomes available once the
detection model has finished downloading — see "Requirements and first run".
YOU STAY IN CONTROL
Nothing is rewritten silently. Every detection is offered to you in a review
panel first: accept the suggested stand-in, edit it, or send the original
instead. Nothing is changed until you say so.
REPLIES STAY READABLE
Redaction normally makes a reply impossible to follow — every name becomes the
same placeholder. The extension keeps a local alias map, so a name is consistently
"Alex Rivera" throughout the conversation and you can still read what came back.
AUDIT LOG
See exactly how many items of each kind were caught, and when. Export it to CSV
for your own records. The log records categories and counts — never the values
themselves.
ATTACHMENTS TOO
Files you attach are scanned as well, not just the text you type. Common document, spreadsheet and plain-text formats are read locally and put through the same detection before the upload goes out.
WHY YOU CAN TRUST IT
• No backend. No account. No sign-up.
• No analytics, no telemetry, no tracking.
• Your prompt text and files never leave your device.
• Detection runs locally: fast regex rules plus an on-device ML model.
• No PiiI-operated servers exist — there is nothing on our side to store, sell or leak.
• Open source under the MIT licence — read every line, or build it yourself.
The one exception is disclosed in full: on first run the extension fetches its
name/address detection model once from the Hugging Face CDN, then caches it.
That request contains none of your data. After it completes, detection is fully
offline.
The extension asks for five permissions and no host permissions at all. It reads
a page only on the chat sites listed above, and never requests access to tabs,
your browsing history, or other websites.
REQUIREMENTS AND FIRST RUN
• Chrome 120 or later.
• On first run it downloads roughly 300 MB of model weights once and caches
them. On a fast connection this takes well under a minute; on a slow or metered
connection it takes considerably longer, and the popup shows the progress.
• Until that download completes, pattern-based categories work and names and
addresses are not yet detected. Press Retry in the popup if it fails.
KNOWN LIMITATIONS
PiiI is a safety net, not a guarantee. It surfaces what it finds and leaves the
decision to you. It cannot promise to catch every sensitive value, and it should
not be relied on as your only safeguard before sending data to a third party.
• Detection is best-effort. Pattern rules and a machine-learning model will
occasionally miss a value, or flag something harmless.
• Chat sites change their page structure often. If interception stops firing on
a site, reload the tab so the extension can re-attach, and check for an update.
• The interface is English only. Detection itself is multilingual.
DELETING YOUR DATA
Pause protection, clear the audit log, clear the alias map and clear the
whitelist at any time from the toolbar popup. Uninstalling the extension deletes
everything it stored, because all of it lives inside your browser profile and
nowhere else.
TROUBLESHOOTING
The popup says the model is unavailable. Pattern rules still work; only names and
addresses are affected. The model is fetched once from a public CDN, so a
corporate proxy or firewall that blocks that CDN will keep it unavailable.
Nothing is detected on a site that should work. Open the popup and check that the
header reads Active. If it reads Paused, click Resume protection, then reload the
tab.
A term keeps getting flagged. Add it to the whitelist in the popup, or select it
on the page and press Alt+Shift+A.
LINKS
Source code:
https://github.com/JaySmith502/PiiI
Full supported-site list:
https://github.com/JaySmith502/PiiI-public/blob/main/SITES.md
Privacy policy:
https://jaysmith502.github.io/PiiI-public/privacy.html
Support and bug reports:
https://github.com/JaySmith502/PiiI-public/issues
PiiI has no backend, no account and no telemetry - see the privacy policy above
for the full detail. PiiI is open source under the MIT license.