Paste text and strip the personal details out of it right in this tab. Nothing is uploaded - every replacement happens on your device.
Example result (what a full clean looks like - no download needed):
[NAME] emailed the quarterly report to [NAME] at [ORGANIZATION] before their meeting in [LOCATION]. You can reach her at [EMAIL] or [PHONE]. She paid the invoice with card [CARD] from account [IBAN].
The emails, phone, card and account number are caught by built-in patterns with no download; the names, company and city are found by the on-device AI model.
Three steps, all in your browser:
- Paste the text you want to share into the box.
- Click "Remove emails, phones & numbers" for an instant clean with no download - it catches email addresses, phone numbers, card numbers (checksum-verified) and IBAN bank accounts.
- Optionally click "Also remove names & places with AI" - it downloads a ~105 MB language model once (cached by this browser for next time), then replaces person names, organizations and locations on your device.
Text Anonymizer - Remove Names, Emails & Numbers (No Upload)
Strip personal details out of any text before you share it - entirely in the browser, nothing is uploaded. The tool replaces the sensitive parts with readable placeholders like [NAME], [EMAIL] and [PHONE] so the text still makes sense afterwards.
Two layers do the work: an instant pattern layer that needs no download, and an optional AI layer for the names no pattern can list.
Key features
- The instant layer finds email addresses, phone numbers, payment card numbers (verified by checksum so ordinary long numbers are left alone) and IBAN bank accounts - with deterministic patterns and no download at all.
- The optional AI layer finds person names, organizations and cities that no pattern can list, and replaces them with [NAME], [ORGANIZATION] and [LOCATION].
- The recognition model (about 105 MB) downloads once on your explicit click, with the size stated first, and is cached by the browser for next time.
- Pattern redaction runs before the AI pass, so card and account digits are already gone before the model reads the text.
- The result panel shows a per-category count of what was replaced and a copy button for the cleaned text.
- If the AI model cannot load, the instant pattern clean still runs - the page is never dead.
When to use it
Paste a support ticket, a log excerpt, an email thread, or a document you want to hand to a chatbot, and let the detection do the work for you. This page REPLACES the personal details it finds; it is different from the find and replace tool on this site, where you type each phrase to swap yourself. Use find-and-replace when you already know the exact strings to remove; use this page when you want the detection done for you. For the photo equivalent of the same job, the face blur tool hides faces in images on-device.
Privacy
Keeping the processing local is the whole point of the task: text you are anonymizing is exactly the text you least want to send to a server. Both layers run on your device - the only network transfer is the one-time model download itself, which contains no part of your text.
What to expect from detection
Review the result before sharing: the AI layer works at a confidence threshold, so it finds most names, not provably every name - unusual spellings, nicknames or names that double as common words can slip through, and English text works best. The instant layer only catches the pattern types listed above; account usernames, street addresses or national ID formats it does not know will remain. This is a practical cleanup aid, not a certified de-identification or compliance tool.
When the sensitive text lives inside a PDF rather than a paste-able block, the Redact PDF tool does the visual equivalent: draw boxes over the lines and download a rebuilt file where the covered content is deleted, not hidden. It uses the same on-device detection model as this page, so the one-time download is shared between the two.
Frequently Asked Questions
Does my text get uploaded anywhere?
No. Both cleaning layers run in your browser on your device. The only network transfer is the optional one-time AI model download (about 105 MB), which contains no part of your text.
What does the instant mode remove without any download?
Email addresses, international-format phone numbers, payment card numbers (only when they pass the Luhn checksum, so ordinary long numbers are kept) and IBAN bank account numbers. Each match is replaced with a placeholder such as [EMAIL] or [CARD].
What does the AI mode add?
Person names, organizations and locations - things no fixed pattern can list. It runs a named-entity recognition model on your device and replaces what it finds at or above a confidence threshold with [NAME], [ORGANIZATION] or [LOCATION].
Will it catch every name?
No, and the page does not claim to. The model finds most names in ordinary English text, but unusual spellings, nicknames or names that double as common words can slip through. Always review the output before sharing it.
Is this a certified de-identification or compliance tool?
No. It is a practical cleanup aid for text you are about to share. It gives no legal guarantee that the output meets any regulation, and it does not know street addresses or national ID formats beyond the pattern types listed on this page.
Which languages work?
The instant pattern layer is language-independent. The AI name detection works best on English text; names in other languages are found less reliably.