Sovereign Shield — LLM PII guard

Redacts Swiss, EU & international identifiers before ChatGPT/Gemini/Claude, then restores them in the reply — 100% local.

Sovereign Shield keeps Swiss, EU and international identifiers out of the big chat assistants. Before your prompt leaves the browser for ChatGPT, Gemini or Claude, it replaces any checksum-validated identifier with a neutral placeholder, then restores the real value in the reply — so the conversation reads normally while the sensitive number never leaves your machine. Everything runs locally. No account, no API key, no server, no analytics, and no data ever leaves your device. Detection is deterministic: an identifier matches only when its regex shape AND its real check digit agree, so ordinary text is never touched and there are no false positives. What it detects: • Swiss AHV/AVS, IBAN (worldwide), credit card • Italy Codice fiscale, Spain DNI/NIE, France NIR, Netherlands BSN • Germany Steuer-ID, Poland PESEL, Portugal NIF, Belgium Rijksregisternummer • UK NHS number • Brazil CPF & CNPJ, South Africa ID, China resident ID, Canada SIN, India Aadhaar You stay in control: choose which categories to block, and an on-device activity log shows what was kept local — type, time and site only, never the value. Before you send, a live count above the chat box shows what will be kept local. The value-to-placeholder map lives only in your tab's memory and is never stored or transmitted. Structured identifiers only — names and addresses are out of scope. Tip: on Gemini, if a message doesn't send on the first Enter right after pasting, just press Enter again — that's Gemini's own editor, not the extension. Open source: https://github.com/acoseac/sovereign-shield