The questions we hear most often.

No. Do Not Feed AI works locally. It does not send data to the vendor, collect statistics or track user activity. It needs no account and no server connection to verify the license. We only receive what you send us yourself: a message from the form or an e-mail. The only connection the extension makes to our website is the "check for updates" button in the extension panel, and only after you click it: it then fetches the updates.json file (with no version number or identifiers), and the extension checks nothing by itself in the background.

We don't promise that. Detection is based on rules and checksums, so an unusual format, a typo or data in an unexpected place can be missed. The AI receives replacements for what the program detected, and the rest of the text goes as it is. That is why there is a preview before sending: it shows the list of detected values and the exact text the AI will receive, so you can fix it if needed or add a missing name as a custom phrase.

Yes, but only locally and without content. The program keeps an audit log on your computer: when it ran, the license state, how much data of which kind was masked and why a file was stopped. The log contains no messages, detected values, names of people or companies, custom phrases or file names. It is protected against unnoticed changes (a hash chain), kept for 180 days and never sent anywhere: we get it only if you send it with a support request. It works with every license, and an administrator has the commands dnf audit verify and dnf audit export.

No, it is pseudonymization (reversible PII masking). Anonymization is irreversible, while the program has to restore the originals in the chat reply, so it keeps the replacement map on your computer. Pseudonymized data is still personal data under the GDPR: the tool reduces the data sent to the chat, but it does not remove the controller's obligations and does not replace your company's rules for using AI. Details: security and the law.

The extension replaces data at the moment of sending. The field you type in belongs to the chat page, though, so the page has technical access to it before you press Enter, for example for suggestions or saving drafts. Very sensitive data is best prepared beforehand, pasted ready and sent right away, rather than typed at length in the chat field.

You add project, client or product names as custom phrases (in Business, the company dictionary is in the license file). Preview mode shows a list of detected values with checkboxes before every send, and manual restore mode shows exactly what the AI received and replied. Known limitation: a person not seen before and mentioned only in an inflected form ("list do Nowaka") is not detected.

Restoring recognizes variants of a replacement: a change of letter case, the order "Surname First name" and "Surname, First name", the short form "J. Surname", other separators in NIP (tax ID), IBAN and phone numbers (also without +48) and Polish case inflection. When the model writes "Zalewskiemu", you will see the original surname on screen, in the same case.

Partly, and it is best to test it on your own samples. Some detection is universal: e-mail addresses, payment cards (checked by checksum), IBANs, IP addresses and host names, and passwords, keys and tokens in over 220 provider formats. Company names with endings such as Ltd. or GmbH are recognized too. The rules are tuned for Polish data, though: national ID numbers (PESEL), tax and business IDs (NIP, REGON), Polish name forms and Polish address and phone formats. Names, addresses and ID numbers from other countries may be caught less reliably or missed, and the custom phrases list lets you add clients, projects and products yourself. If you work with data from another country, ask for a trial license and test the program on your own examples; we can tell you what is and is not covered.

Yes. All the detection, replacements, restoring replies and the license live in the program on your computer; the extension only passes text and files to it. Without the program the extension does not mask data: the red icon in the chat asks you to install it, and before sending a message it asks whether to send it without masking. The program runs while the browser is open. On Windows one installer sets up the program and the extension at once.

First install the program (the installer for Windows or macOS is in the Installation section), then add the extension from the store: in Chrome click Add to Chrome. The same link works in Edge, which may ask for permission to use extensions from other stores: confirm it. Firefox: the extension page in the Mozilla store, Add to Firefox, then Add. If a company network blocks the store, there is also a manual install for IT, described in the Installation section.

Add to ChromeAdd to EdgeAdd to Firefox

The program is rules and checksums, with no language models and no cloud: checking a few thousand characters takes a few milliseconds, so the number of detected items on the chat icon updates live as you type. The replacement in the field happens at the moment of sending, and then the message goes on as usual.

Windows 10 and 11 (also Windows 11 on ARM): Chrome, Edge and Firefox. macOS 14 or newer: Chrome and Firefox (Edge on macOS is not supported). We mask data in the ChatGPT, Claude, Gemini and Copilot chats. We do not support other browsers, other systems or mobile browsers.

Yes. The extension is in the Mozilla store: open its page in Firefox, click Add to Firefox, then Add. It stays installed for good, without developer mode, and updates itself. The program on your computer is the same as for Chrome and Edge. Supported browsers: on Windows, Chrome, Edge and Firefox; on macOS, Chrome and Firefox.

Yes, but in a different way than in the browser. The extension does not work inside an editor or a terminal, but the program has clipboard shortcuts (a Pro and Business feature, off by default, turned on in the program's menu icon). Windows: Ctrl+Win+A (Anonymize) and Ctrl+Win+R (Restore); macOS: Ctrl+Option+A and Ctrl+Option+R. You copy text in any program, the shortcut turns the clipboard into text with replacements, you paste it into the AI, then you copy the reply and restore it with the second shortcut. The shortcut does not watch what you type: the clipboard is read only after you press it. We do not support editor assistants or API calls. The system clipboard history (for example Win+V) and clipboard managers may keep the original that was copied before you used the shortcut, so turn them off while you work with client data.

The program's menu icon (the menu bar on macOS, the tray on Windows) has the commands "Anonymize file..." and "Restore file...": txt, csv, xlsx and docx, and also a PDF with a text layer, which we turn into a docx or txt (without page layout, images, hyperlinks and signatures; scans without text are rejected). From jpg and png images the program removes only the metadata, such as location, device or author, and the content of the image (text in a photo, faces) is not masked. The result gets a neutral name ending in _anon, the original stays unchanged, and the map of originals is stored encrypted on your computer for 30 days. The feature is in Pro and Business. In the chat, PDFs and images are still blocked.

Request a trial license with the form on the home page. We e-mail you a 7-day license file with all the Pro features, with no account and no commitment. You can download the program and the extension right away, but without a license file protection is off.

We don't publish prices: each offer is prepared for the number of seats and the way you deploy. Write to us with the form and we will reply with a proposal for your company.

The extension warns you 30 days in advance: a badge on the icon, a notice in the panel, and in the last 14 days also an amber ring around the chat icon and a notice in its tooltip and menu. After the expiry date you still have 14 days of full access (a grace period). After that protection is off: the extension does not change messages or files, and the red icon in the chat says so plainly. Replies in open tabs still come back with the originals. A trial license works the same way. Details: what works after expiry and how to renew a license.

The extension is in the Chrome Web Store, so in Chrome you can deploy it with the ExtensionInstallForcelist policy and the store ID; in Edge and on networks that block the store, use the manual install from our package. The program goes out via MDM or a silent per-user installer (Windows: /VERYSILENT), without administrator rights. On Windows there is also an MSI installer for IT departments (Group Policy, Intune, SCCM; per-computer or per-user install). MDM can place the donotfeedai.lic license file in the data folder, so users do nothing, and the company dictionary from the license is added to the custom phrases on every workstation.

Shadow AI is using AI chats at work outside the company's rules, for example from a personal account: an employee pastes client data into a chat and the company does not know. It affects any company where people have access to AI chats. Do Not Feed AI does not replace a policy or training, but it reduces the client data that reaches the chat, whichever account the employee uses.