Transform

masking_type: "transform" — Custodian Labs' proprietary de-identification. It swaps sensitive values for realistic stand-ins and keeps the sentence readable, so the output still works for LLMs, demos, testing, and downstream processing.

Python

from custodian_labs import GuardianLayer

result = GuardianLayer().deidentify_text_outputs(
    "Dr. Emily Chen called Alex at 555-0100 about the appointment in Auckland.",
    masking_type="transform",
    pii_entities=["PERSON", "PHONE_NUMBER", "LOCATION"],
)
for item in result.outputs:
    print(item.id, item.text)

TypeScript

import { GuardianLayer } from "@custodianlabs/sdk";

const result = await new GuardianLayer().deidentifyTextOutputs(
  "Dr. Emily Chen called Alex at 555-0100 about the appointment in Auckland.",
  { maskingType: "transform", piiEntities: ["PERSON", "PHONE_NUMBER", "LOCATION"] },
);
for (const item of result.outputs) console.log(item.id, item.text);

outputs holds two candidates. Use outputs[0] by default; show both if a reviewer picks.

1  Dr. Sarah Miller called Jordan at 555-0188 about the appointment in Wellington.
2  Dr. Laura Adams called Casey at 555-0142 about the appointment in Christchurch.

Options

Option	Notes
masking_type	"transform" (default) or "redact".
pii_entities	Categories to process; omit or ["ALL"] for everything. See Supported Entities.
replacements	{ "original": "replacement" } to pin specific swaps (on deidentify_text).

Transform vs Mask

	transform	redact
Output	Dr. Sarah Miller called Jordan...	Dr. ***** called ***** ...
Readable	Yes	No
Best for	LLM input, sample data, demos	Audit, "prove it's gone"

See also

Masking · File De-identification