Media
5 items
Description
Detect and redact PII in your ChatGPT prompts before you send them. Runs entirely in your browser — no server, no API key. Strip personal info out of your ChatGPT prompts before you send them — runs entirely in your browser. Rescriber scans the message you're about to send to ChatGPT and flags any personally identifiable information (PII) it finds — names, email addresses, phone numbers, postal addresses, URLs, dates, account numbers, passwords, and API keys. With one click you can replace each item with a placeholder like [NAME1] before the message leaves your browser. When ChatGPT's reply references [NAME1], the original value is shown back to you in-place. The model only ever sees the placeholder. Why install it Every PII detail you include in a prompt becomes another piece of your life sitting in someone else's logs. Rescriber gives you a chance to spot and redact those details before they get there — so you can ask ChatGPT for help with real-world tasks (drafting an email, debugging code that contains a key, summarizing a medical question) without handing over more about yourself than you meant to. How it works Rescriber loads a small open-source PII-detection model (Hugging Face's openai/privacy-filter, ~30–50 MB) into your browser the first time you use it, then runs it locally on every prompt — using your GPU when available, with a CPU fallback. Your prompts and the detected PII never leave your computer. The only outbound network request the extension ever makes is to download the model file from Hugging Face on first use. What it stores Per-conversation mappings between your real PII and the placeholders you chose are saved locally in chrome.storage.local so the originals can be restored when ChatGPT's reply references them. You can view, edit, and delete every piece of stored data from the extension's options page. No analytics, no telemetry, no third-party trackers. Limitations Works on chatgpt.com only. The on-device model is fast but isn't perfect — review its suggestions before redacting. First-time use needs an internet connection to download the model file; detection runs offline thereafter. Open source Source code (MIT) and full privacy policy: https://github.com/PEACH-Research-Lab/Rescriber. Based on the CHI '25 paper "Rescriber: Smaller-LLM-Powered User-Led Data Minimization for LLM-Based Chatbots" by the PEACH Lab at Northeastern University. Rescriber is an independent project and is not affiliated with, endorsed by, or sponsored by OpenAI.