Guardian Layer in Your Stack
Guardian Layer sits between sensitive input and whatever consumes it:
user input → your backend → Guardian Layer → de-identified output → LLM / storage / another service
Call it from your backend, not the browser — the API key is a secret, and your server should own logging, retries, and usage limits.
Pick a call
| You have | You want | Call |
|---|---|---|
| Text | See what's sensitive first | analyze_proprietary() |
| Text | Readable de-identified text | deidentify_text_outputs() with transform |
| Text | Values visibly gone | same, with redact |
| A file | A de-identified file back | deidentify_file() |
| An agent that masks automatically | PII stripped before the LLM, restored after | privacy_enabled on a Custodian Agent |
Inline masking during chat
privacy_enabled=True (Python) / privacyEnabled: true (TypeScript) on a Custodian
Agent or team runs the same de-identification server-side on every message before the
LLM, then restores the original values in the reply. No extra call from your code.
See also
Guardian Layer Overview · File De-identification · Guide: Use PII masking with LLMs