Redact Text
Hide names, emails, phone numbers, addresses, card numbers and API keys before you paste text into ChatGPT or share it anywhere. Swap each one for a label like [NAME_1] or black it out, then put the real details back into the reply. Runs entirely in your browser, no upload, no signup.
Runs on your device. Your text is never uploaded.
Highlights show up as you paste.
Emails, phones, cards and keys are caught already. Names, addresses and places need a small AI model that runs on this device.
28.1 MB download, once. Your text still stays on this device.
Your text never leaves this device. Detection runs in your browser; nothing is uploaded or logged. Free, unlimited, no signup.
Redacted text
Your redacted text shows up here, ready to paste into ChatGPT or anywhere else.
The same value always gets the same label, so the reply still makes sense. Click a hidden value to keep it as is.
Check before you share. Automatic detection can miss things.
Got a reply from ChatGPT? Put the real details back
The labels and their real values live only in this page's memory. Nothing is saved, and a reload forgets them.
Get the next tool first
New local-AI tools and updates by email. Unsubscribe anytime.
Redact text before you paste it into ChatGPT
Emails, support chats, contracts and meeting notes are full of details that are not yours to share: names, phone numbers, addresses, account numbers, the odd API key. This tool finds them, hides them, and hands you a copy that is safe to paste into ChatGPT, Claude, Gemini, a forum post or a bug report.
Paste your text and the highlights appear as you type. Switch whole categories on or off, click a single value to keep it, add your own words, then copy the redacted version. Detection happens on your device, so the original never travels anywhere.
How to use it
- Paste the text you want to share, or try the example.
- Press Also find names and addresses if the text mentions people or places. The first time, this downloads a 28.1 MB model.
- Check the highlights. Switch categories off, click any value you want to keep, and use Redact a word for anything that was missed.
- Pick Labels for AI chats or Blocks for documents, then copy or download.
- When the reply comes back, paste it under Got a reply from ChatGPT? to put the real details back.
How it works on your device
Two layers do the finding. The first is a set of rules for anything with a fixed shape: email addresses, phone numbers, IP and MAC addresses, links, dates, and the key formats of services like OpenAI, Anthropic, GitHub, AWS, Stripe and Slack. Numbers are checked, not just matched: a card number has to pass the Luhn checksum and an IBAN the bank checksum, so a random order number is not mistaken for either. The rules run instantly on every keystroke.
The second layer is a small AI model that reads the text the way a person would and picks out names, street addresses, cities and organisations, which no rule can describe. It runs inside your browser on your own processor. Where the two layers disagree, a checked number wins over the model, and the model wins over a loose pattern, so a surname is never mistaken for a phone number.
Labels that keep the meaning
A chatbot can only help with text it understands. Blacking everything out leaves it guessing; numbered labels do not. With Labels, "Priya emailed Daniel, then Priya called the bank" becomes "[NAME_1] emailed [NAME_2], then [NAME_1] called the bank". The AI still knows there are two people and who did what, without ever learning who they are. When it answers, the restore box turns [NAME_1] back into Priya.
Need a redacted text generator for a meme, a screenshot or a document? Switch to Blocks and every value becomes a solid black bar of the same length, line breaks intact.
What it does not do
It does not anonymise data in a legal sense and it is not a compliance tool. A detail that only identifies someone in context, like "the only nurse on the night shift", has no shape for a rule and no name for a model to find. Read the result before you share it; you know your text better than any detector.
Private by design, free for good
Online redaction services ask you to upload the very text you are trying to protect. Here nothing is uploaded: the rules and the model both run in your tab, and the page works offline once the model is cached. We have no server bill for the detection, so the tool is free, unlimited and needs no account.
Frequently asked questions
- Is my text uploaded anywhere?
- No. Detection runs in your browser tab, on your device. The text you paste, the values it finds and the redacted result are never sent to a server, never logged and never saved. Close the tab and they are gone.
- What does it find?
- Emails, phone numbers, card numbers (checked with the Luhn checksum), IBANs (checked with the bank checksum), US Social Security numbers and other ID numbers, IP and MAC addresses, links, dates, and API keys or passwords from well-known services. Turn on name detection and it also finds people's names, street addresses, cities and organisations.
- Why is name detection a separate download?
- Names have no fixed pattern, so they need a small AI model rather than rules. It is a 28.1 MB download the first time you switch it on, kept in your browser afterwards so later visits start without downloading anything. Everything else on the page works without it.
- Can I trust it to catch everything?
- No tool can promise that, and this one does not. The rules are exact for things with a fixed shape, like emails, card numbers and keys. Names and addresses come from a model, and models miss some, especially unusual names, names written in lower case and names in languages other than English. Read the output before you share it, click any value you want to keep, and add anything it missed with Redact a word.
- What is the difference between Labels and Blocks?
- Labels swaps each value for a placeholder like [NAME_1] or [EMAIL_2]. The same value always gets the same label, so a chatbot can still follow who said what and reply sensibly. Blocks blacks each value out with █ characters of the same length, the classic redacted document look, which suits screenshots and documents shared with people.
- How do I put the real names back into ChatGPT's reply?
- Keep the tab open, open "Got a reply from ChatGPT?" under the output and paste the reply. Every label the tool handed out, like [NAME_1], is swapped back for the original value. It also understands the variations chatbots tend to write, such as [Name 1]. The labels and their values only exist in this page's memory, so a reload forgets them.
- What is redacted text?
- Redacted text is text with sensitive parts removed or blacked out before it is shared. Governments and courts black out names and numbers in released documents; companies redact customer details before passing records on. Online, the same idea protects you when you paste an email, a support chat or a document into an AI chatbot or a public forum.
- Does it work in other languages?
- The rules work in any language, because an email, an IBAN or a card number looks the same everywhere. The name model was trained on English, so names in other languages and scripts are found less reliably. Use Redact a word for those.
- Does it work on my phone?
- Yes. The rules run instantly on any phone. Name detection is a one-time 28.1 MB download that then runs on the phone itself; switch it on the same way as on a computer.
- Is it really free? What is the catch?
- Free, unlimited and no signup. There is no server doing the work, so there is no cost per use to pass on to you. The model is Gravitee's bert-small-pii-detection, released under the Apache 2.0 licence, and your browser downloads it straight from Hugging Face.
- Can I redact a PDF or an image?
- This page works on text. For a screenshot or a scanned page, pull the text out first with the image to text tool, then paste it here.
Pairs well with
- Image to text: pull the text out of a screenshot or a scan, then redact it here.
- Audio transcription: turn a recorded call into text on your device before you share a summary.
- Word counter: check how many words and tokens the redacted text uses.