Anonymize for AI: text, PDF and Word files, on your device (KeptPDF)

Anonymize text for AI on your device, no uploads. Reads a page only when you ask, or on a site where you turn on auto mode.

Before you paste into an AI chat, KeptPDF finds the private details in your text: names, Social Security numbers, phone numbers, emails, addresses, account numbers, organizations and dates. It swaps each one for a placeholder like [PERSON_1], so the AI sees the placeholder in place of each detail it found. It can miss names, so check the result. Open the side panel next to your chat, paste your text, and copy the clean version. When the AI answers, paste its reply back and put the real names back. You can also drop in a PDF, a Word file, a text file or a picture. Scans and pictures are read on your device. Or work right inside the chat box. Select your text, right-click, and choose KeptPDF > "Anonymize here". The placeholders replace it on the spot. When the answer comes back, select it, right-click, and choose "Put real names back". Shortcuts: Alt+Shift+H anonymizes the box you are typing in, and Alt+Shift+U puts real names back in the text you selected. You can change them at chrome://extensions/shortcuts. A short note appears on the page to tell you what happened. Need to share a screenshot? Press Alt+Shift+S, or right-click the page and choose KeptPDF > "Clean screenshot". The side panel shows a picture of the part of the page you can see, with the names, numbers, faces and barcodes it found already covered in black. Click a box to uncover it, drag on the picture to cover more, then copy the image or save it. The picture is made and covered on your device, and what you save has the boxes painted into its pixels. It can miss things, so check it before you share it. Chrome does not allow a picture of its own pages, such as chrome:// pages or the Chrome Web Store. Or turn on auto mode. It is off until you turn it on, one site at a time: ChatGPT, Claude, Gemini or Copilot. While it is on for a site, pressing send there anonymizes your message first. A file you paste, drop or pick in the chat is held until you choose. "Anonymize, then add" reads it on your device and adds a clean text file with placeholders. The AI gets the text only: layout and pictures are left out. Or you can add the original, real names included. Three limits to know. The site can read what you type before KeptPDF changes it. A file picker the page makes by itself and never puts on the page can reach the site without a note. And KeptPDF can miss names, so read your message before you send. A saved list on your device keeps the same person on the same placeholder across messages. Name your lists, switch between them, export one, set it to forget lists you have not used for 1, 7 or 30 days, or delete everything in one click. Private by design. It runs on your device. Nothing is uploaded. At install the extension has no access to any site. Otherwise it only sees text you paste, text you select and choose to anonymize, or, when you ask for a clean screenshot, a picture of the visible part of that tab. If you turn on auto mode for one of four AI chat sites, it can read that site's pages while it is on. Auto mode can miss names, does not cover voice input, and the site sees what you type before it is anonymized. One thing to know: "Put real names back" in the chat box writes the real names into the page you are on, so that site can see them, the same as if you had typed them. To keep them off the site, use Put real names back in the side panel instead. Want to redact a PDF? A link in the panel opens KeptPDF on the web. New in 1.2 - Clean screenshot: a picture of the visible tab with names, numbers, faces and barcodes covered. Uncover or add boxes, then copy or save it. Alt+Shift+S, or right-click > KeptPDF > Clean screenshot. - Auto mode: add a PDF, Word file, text file or picture in the chat and choose "Anonymize, then add". It is read on your device and added as clean text with placeholders. Layout and pictures are left out, apart from words KeptPDF can read in a picture on a PDF page. The note counts the pictures that were not read, and says that headers and footers in Word files are not read. Text boxes are read, unless Word draws them only as a shape. It can miss names. The original is only added if you choose it. - Finds a name after a word like Patient, Client or Tenant. - Keep the layout: a stopped PDF or Office file can be opened in Redact on keptpdf.com. The extension hands the file over inside your browser, and Redact runs in your browser, so nothing is uploaded. - Anonymous counts of actions and errors, on by default, with an off switch.