dfki-dsa

dfki-dsa / pii-guardrail-browser-extension

Public

Privacy Guardrail is a local-first Chrome extension that detects and anonymizes PII before text is pasted into ChatGPT, Claude, or Gemini, using Rust/WASM recognizers and optional in-browser transformer NER.

14
4
100% credibility
Found May 22, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Privacy Guardrail is a Chrome browser extension that helps protect your personal information when using AI chatbots like ChatGPT, Claude, and Gemini. When you try to paste text containing sensitive data (email addresses, phone numbers, credit cards, names, etc.), the extension pauses the paste and shows you exactly what personal information was detected. You can then approve, edit, or cancel the paste. Approved text gets anonymized with placeholders before sending, so the AI never sees your real personal data. Everything runs locally on your computer—nothing is sent to external servers. The extension also includes a clipboard feature that lets you restore anonymized text back to its original form later if needed.

How It Works

1
🔍 You hear about Privacy Guardrail

A friend tells you about a browser extension that stops personal data from accidentally going into AI chatbots.

2
📦 You install the extension

You add Privacy Guardrail to Chrome from the web store. It asks permission only to work on the AI chat sites you use.

3
✨ You paste something with personal info

You copy a message containing your email and phone number, then paste it into ChatGPT. The extension catches it first.

4
👀 A friendly review screen appears

Before your text goes anywhere, a small window shows what personal information was found and asks you to confirm.

5
You choose what to share
✅
Approve and paste

Your text goes to the AI, but with placeholders like [EMAIL_1] instead of your real email.

✏️
Edit the selection

You uncheck items you don't want anonymized, like a company name that's not sensitive.

🚫
Cancel the paste

Nothing gets sent. You can rewrite your message without personal data.

6
🤖 The AI responds about your anonymized text

The AI sees [EMAIL_1] and [PHONE_1] and responds accordingly. Your real contact info never left your computer.

🔓 You can restore originals later

If the AI mentions your anonymized data, you can copy it back and the extension restores your real values from your personal vault.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is pii-guardrail-browser-extension?

Privacy Guardrail is a Chrome extension that intercepts text before you paste it into ChatGPT, Claude, or Gemini, scans for personally identifiable information, and replaces detected PII with stable placeholders like [EMAIL_1] or [PERSON_1]. It runs detection entirely in your browser using a combination of regex-based pattern matching and optional transformer-based NER, with no text ever sent to a remote server. The placeholder map stays local so model responses can be de-anonymized later with the original values highlighted.

Why is it gaining traction?

The privacy-first architecture is the main draw. Everything runs client-side with zero telemetry, which matters for developers working with sensitive data. The two-layer detection approach is also more thorough than simple regex: structured formats like credit cards and IBANs are validated with checksum algorithms (Luhn, mod 97), and context keywords boost confidence when "email" appears next to an email address. Supporting both English and German adds practical value for EU developers.

Who should use this?

Developers in regulated industries (healthcare, finance, legal) who use AI chat tools and need a local safety net against accidental PII leakage. Technical users comfortable with beta software who have machines meeting the higher system requirements (16GB RAM minimum, WebGPU-capable GPU recommended for smooth Local AI). Not yet suitable for teams needing mobile support, cross-browser compatibility, or compliance certifications.

Verdict

This has a 1.0% credibility score reflecting its early stage: only 14 stars, public beta status, and limited user base. The Rust/WASM detection engine is well-engineered with solid test coverage, and DFKI's institutional backing adds credibility. However, the high system requirements, narrow browser support, and beta maturity mean it's not production-ready for most teams. Worth watching and experimenting with, but don't bet sensitive workflows on it yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.