Media
4 items
Description
Catches personal data before you send it to an AI chat. The check runs locally; the extension never sends or stores what you type. CloakLLM Guard checks what you are about to send to ChatGPT, Claude, Gemini or Microsoft Copilot, and stops you if it contains personal data. Paste a customer record into a chat box and press send, and instead of the message going out, Guard stops and asks: "Hold on - this looks like personal data. What you are about to send appears to contain a credit card number and an email address." You choose "Let me edit it" or "Send anyway", and you can always send. It is a seatbelt, not a lock. THREE THINGS IT NEVER DOES - Send anything anywhere. There are no network requests at all. No server, no analytics, no account. - Store what you typed. It records that a warning happened and which kinds of data were found - never the values, never the surrounding text. - Report to anyone. It warns you. Your employer does not get a copy. Detection runs entirely in your browser, using the detection engine from CloakLLM, an open-source compliance toolkit built to keep personal data out of AI audit logs. WHAT IT CATCHES Credit card numbers, IBANs, US Social Security numbers, API keys, AWS access keys, access tokens, email addresses and phone numbers. Every category can be turned off individually - if something is noisy in your work, silence it. A warning you learn to ignore is worse than no warning. WHAT IT CANNOT SEE Browser tabs only. It cannot see the ChatGPT or Claude desktop apps, Copilot or Cursor inside your editor, anything sent directly from code, or anything on another device. That is a real limit, and we would rather you knew it now than found out later. It also does not recognise names, postal addresses or dates of birth. It catches the structured kinds of data listed above; a person's name on its own will not trigger a warning. YOUR OWN RECORD Every warning is logged locally - categories and counts only - so you can see whether this is actually helping: how many near-misses, how many you acted on, which kinds keep coming up. Export it or delete it whenever you like. It is your record, not ours; we never see it. Open source, MIT licensed: https://github.com/cloakllm/cloakllm-guard