PII Redaction (PPE)

This Apify Actor redacts PII/PHI/PCI from text using a remote redaction API.

What it does

• Takes either:
  • INPUT.text (a string or array of strings), or
  • INPUT.datasetId (reads items from an existing dataset and redacts selected fields)
• Sends text to a redaction API for entity detection + redaction (marker or mask)
• Makes one API call per input text
• Writes multiple output records:
  • one record per sentence (type: "sentence")
  • one record per entity (type: "entity")

Pricing (PPE)

The Actor uses Pay-Per-Event.

• Event: REDACTION_REQUEST (configurable via chargeEventName)
• Charged per output entry: One PPE event per dataset record pushed.
  • each type: "sentence" record charges once
  • each type: "entity" record charges once

This matches Apify's PPC/PPE design where events correspond to output units.

Input

Parameter	Type	Default	Description
apiKey	String	(optional)	API key (sent as x-api-key). Use an Apify Secret in production.
text	String | String[]		Text(s) to redact.
datasetId	String		Dataset to read items from.
fieldPaths	String[]	["text"]	Field paths (dot/bracket notation) to redact inside each dataset item, e.g. "text", "customer.message", "items[0].note".
split.enabled	Boolean	true	Whether to split the redacted output into sentences for dataset records.
hideit.endpoint	String	(default set in code)	Override the redaction endpoint URL.
hideit.entityTypes	String[]	(sane default list)	Entity types to enable.
hideit.outputMode	String	MARKER	Redaction output mode: MARKER (replace with typed markers like [NAME_1]) or MASK (replace with a fixed token).
hideit.maskToken	String	[REDACTED]	Only used when hideit.outputMode="MASK".
hideit.markerPattern	String	[UNIQUE_NUMBERED_ENTITY_TYPE]	Marker pattern for redaction output.
hideit.markerLanguage	String	en	Marker language.
hideit.coreferenceResolution	String	heuristics	Coreference setting supported by the endpoint.
proxy	Object		Optional Apify proxy configuration.
chargeEventName	String	REDACTION_REQUEST	PPE event name to charge.

Output modes

• MARKER: returns structured placeholders like [NAME_1], [PHONE_NUMBER_1].
• MASK: returns a uniform token like [REDACTED].

Output dataset schema

Two kinds of output records are written:

Sentence records (type: "sentence")

• type: "sentence"
• source: "input.text" or "dataset"
• sourceMeta: metadata like index, or { datasetId, itemIndex, fieldPath }
• sentenceIndex, sentencesTotal
• text (one redacted sentence)
• hideit request metadata (endpoint + mode)
• createdAt

Entity records (type: "entity")

• type: "entity"
• source, sourceMeta
• entityIndex, entitiesTotal
• entity (as returned by the endpoint)
• createdAt

Notes

• Keep your API keys out of git. Prefer Apify Secrets.
• For very large datasets, you may want to implement pagination + concurrency. Current implementation reads the dataset first page only.

Length limits

• limits.maxInputChars (default: 10000): if an input text exceeds this length, the actor logs max reached, truncates the text, and continues processing.