Rampart Redact

The rampart_redact function detects and redacts Personally Identifiable Information (PII) in a piece of text locally, using the bundled Rampart ONNX model (no external call). Use it as a standalone node in a workflow to scrub text before sending it elsewhere, or to inspect what PII a payload contains.

• Function name: extensions.com.cloud-apim.llm-extension.rampart_redact

Parameters

Parameter	Type	Required	Default	Description
input	string	yes	—	The text to scan / redact
action	string	no	redact	redact (return redacted text + mapping) or detect (return spans only)
min_score	number	no	0.4	Confidence floor for the model
entities	string[]	no	standard PII set	Entity types to consider

See the Rampart guardrail for the full list of entity types and how the deterministic + model layers work.

Output

For action: "redact":

{
  "redacted": "My name is [GIVEN_NAME_1] [SURNAME_1] and my SSN is [SSN_1].",
  "mapping": {
    "[GIVEN_NAME_1]": "Alex",
    "[SURNAME_1]": "Rivera",
    "[SSN_1]": "472-81-0094"
  },
  "spans": [
    { "entity": "GIVEN_NAME", "start": 11, "end": 15, "score": 0.99, "text": "Alex" }
  ]
}

For action: "detect":

{
  "found": true,
  "count": 3,
  "spans": [
    { "entity": "SSN", "start": 37, "end": 48, "score": 1.0, "text": "472-81-0094" }
  ]
}

Example

{
  "kind": "rampart_redact",
  "function": "extensions.com.cloud-apim.llm-extension.rampart_redact",
  "args": {
    "action": "redact",
    "input": "My name is Alex Rivera and my SSN is 472-81-0094."
  },
  "result": "redaction_result"
}

A common pattern is to redact first, call an LLM on redacted, then re-inflate downstream using the returned mapping.