PromptSafe: catch sensitive data in AI prompts

Catches client data, financial data and secrets in AI prompts before you send them. Runs locally.

PromptSafe helps small firms and careful individuals avoid accidentally pasting sensitive data into AI chat tools. When you paste or send a prompt on ChatGPT, Claude, Gemini, or Dola, PromptSafe checks it on your device. If it finds something sensitive, it shows a warning with masked previews and lets you: Redact and continue (replaces values with placeholders like [SSN_1]) Send anyway (warn mode only) Cancel What it detects: Payment card numbers (checksum-verified) IBANs (checksum-verified) US Social Security numbers and labeled EIN/tax IDs Email addresses and phone numbers Dates of birth, passport numbers, bank account numbers (when labeled), US routing numbers (checksum-verified), ITINs AWS keys, GitHub tokens, API keys, private keys Optional packs: medical record numbers; legal matter/case numbers and court dockets; international IDs (UK National Insurance, Singapore NRIC/FIN, Philippine TIN, SSS, PhilSys, PhilHealth); IP addresses Protected names: add client or company names and PromptSafe flags them wherever they appear. The list stays on your device. False-positive controls: mark a flagged value "Not sensitive" and PromptSafe stops flagging that exact value (remembered only as a salted hash on your device), or switch off a rule that keeps misfiring from the options page. AI tool inventory: see which supported AI tools are being used in this browser, with first and last seen dates and prompt counts (counts only, exportable as CSV). Privacy first: Everything runs locally. Prompt text is never sent anywhere. The audit log stores rule names and counts only, never content. No analytics, no remote code, minimal permissions. Good to know: PromptSafe reduces accidental leaks. It is not a guarantee and does not catch every kind of sensitive data (for example, names you have not added to your protected list, or addresses and other free text). File attachments are not scanned yet; you will see a reminder. PromptSafe is built by Umbrella Web (https://umbrellaweb.net/). PromptSafe is an independent product and is not affiliated with or endorsed by OpenAI, Anthropic, Google, or Dola. Product names belong to their owners. Building a team or firm? Join the waitlist for shared policy and an audit report: https://promptsafe.umbrellaweb.net