What to do before a model sees the text
Prompts, logs and tickets copied into an external chat often carry emails, phone numbers, tax identifiers and keys. This tool replaces them with stable placeholders, [EMAIL_1] or [CF_1], and does not send the text anywhere.
The legend stays in the tab so you can check each match. Restore puts a false positive back into the redacted text. When the model replies and keeps the markers, paste that reply into the same tool: legend placeholders become the original values again, still only in this tab.
The Italian page is thesanitizzatore prompt. The wider boundary is on Governed AI.