A startup wants to keep using its preferred LLM while reducing raw customer PII exposure.
User submits natural-language text containing names, email addresses, phone numbers or local identifiers.
Guard().wrap(...) calls PrivoNest detection before the selected model provider.
Detected values are replaced with scoped aliases so the LLM can keep the semantic role without the original value.
The chosen provider produces a response containing aliases where relevant.
The application restores exact matching aliases using short-lived local mapping.
The startup can change LLM providers while keeping the same PrivoNest privacy layer.