Guardian Layer

Guardian Layer is CustodianAI's proprietary detection layer. It identifies domain-specific sensitive content that standard NER-based compliance modes miss — proprietary terminology, internal identifiers, and contextually sensitive language specific to your organization or industry.

Use it when your documents contain sensitive information that doesn't fit standard PII categories.

---

Basic usage

Set compliance_mode to "PROPRIETARY" on the de-identification endpoint:

import requests

response = requests.post(
    "https://api.custodianai.com/api/v1/deidentify/text",
    headers={"X-API-Key": "cai_your_key_here"},
    json={
        "text": "The Q3 results for Project Helix exceeded targets by 18%.",
        "compliance_mode": "PROPRIETARY",
        "domain": "General",
    },
).json()

print(response["output_text"])

---

Parameters

Parameter	Type	Default	Description
domain	str	"General"	Content domain. See Compliance Modes for options
replacements	dict	None	Pre-supplied replacement map. When provided, Guardian Layer applies it directly without running detection

---

Output modes

The masking_type parameter on the /deidentify/text/proprietary/outputs endpoint controls what detected terms are replaced with:

Value	Behaviour	Example output
redact	Replace with *****	"Project ***** exceeded targets"
transform	Replace with a plausible alternative that preserves readability	"Project Falcon exceeded targets"

redact is stricter; transform produces output that reads naturally and is harder to identify as redacted.

---

Entity filtering

By default, Guardian Layer targets all detected sensitive content. Use pii_entities to limit detection to specific categories:

response = requests.post(
    "https://api.custodianai.com/api/v1/deidentify/text/proprietary/outputs",
    headers={"X-API-Key": "cai_your_key_here"},
    json={
        "text": "Dr. Emily Chen reviewed trial NX-4401 on March 3rd.",
        "domain": "General",
        "masking_type": "redact",
        "pii_entities": ["PERSON", "ID_NUMBER"],
    },
).json()

Pass null or ["ALL"] to target everything detected. See Supported Entities for the full list of values.

---

Alternative outputs

The /deidentify/text/proprietary/outputs endpoint returns two independently de-identified versions of the input text. Useful for review workflows or generating synthetic training data from sensitive documents.

response = requests.post(
    "https://api.custodianai.com/api/v1/deidentify/text/proprietary/outputs",
    headers={"X-API-Key": "cai_your_key_here"},
    json={
        "text": "The Meridian account is managed by Dana Reyes.",
        "domain": "General",
        "masking_type": "transform",
    },
).json()

print(response["outputs"][0]["text"])  # first alternative
print(response["outputs"][1]["text"])  # second alternative

→ API Reference: /deidentify/text/proprietary/outputs

---

Detection only

To run Guardian Layer detection without modifying the text — useful for auditing or building review UIs — use the analyze endpoint:

response = requests.post(
    "https://api.custodianai.com/api/v1/analyze/text/proprietary",
    headers={"X-API-Key": "cai_your_key_here"},
    json={
        "text": "The Meridian account is managed by Dana Reyes.",
        "domain": "General",
    },
).json()

print(response["sensitive_words"])

→ Detection vs. De-identification · API Reference: /analyze/text/proprietary