🛡️ Sovereign Shield

Redacts Swiss, EU & international identifiers before they reach Gemini, ChatGPT or Claude, and restores them in the reply — 100% local.

Guard enabled

Turn the whole guard on or off.

Smokescreen mode

Send realistic stand-ins instead of [EMAIL_1] placeholders — e.g. your address becomes alice.morgan@example.org. Models write much better email drafts, grammar fixes and formatting when the prompt reads naturally. The real values still never leave this page, and are restored in the reply as usual.

Applies to emails and your custom terms only. IDs, IBANs, cards and secrets always use bracket placeholders: a checksum-valid fake AHV or IBAN could be a real person's number, so we never invent one.

What to block

Identifiers are checksum-validated, so they never false-positive. Unchecked types pass through untouched. Names and addresses are out of scope.

Secrets & API keys

Matched by shape (e.g. AKIA…, sk-…, JWTs, PEM keys). Anchored on high-signal prefixes to keep false positives near zero, but — unlike the identifiers above — these are pattern-matched, not checksum-verified.

Custom rules

Redact your own terms — internal code names, client names, domains. Kept local and restored in the reply, exactly like the built-in types. Plain text by default (whole-word); tick Regex for a pattern. Rules never leave your browser.

Activity

Type, time and site only — never the value, not even masked. The value↔placeholder map stays in page memory and is never saved to disk.

Support

Questions, bug reports or feedback — we're happy to help.