presidio Classifier

Defines a classifier that detects personal and sensitive data using a Microsoft Presidio analyzer service.

The presidio classifier sends the text to the analyzer's /analyze endpoint and reports a label as detected when the analyzer finds the corresponding entity, such as EMAIL_ADDRESS or CREDIT_CARD, at or above the configured score threshold. Presidio recognizes content that a fixed expression can't, such as a person's name or a location.

classifiers:
  presidio0:
    type: presidio
    options:
      endpoint: http://presidio-analyzer:3000
      threshold: 0.5
      labels:
        email: EMAIL_ADDRESS
        card: CREDIT_CARD

A classifier is applied to a message value by the classify model. An empty value is never detected. If the analyzer can't be reached, times out, returns an error, or returns a malformed response, the value is treated as detected, so the message is rejected rather than passed through unchecked.

Configuration (* required)

type: presidio

const

Specifies the classifier is presidio.

options*

object

The presidio specific options.

options:
  endpoint: http://presidio-analyzer:3000
  credentials:
    authorization: Bearer ${{env.PRESIDIO_TOKEN}}
  language: en
  threshold: 0.5
  timeout: PT5S
  labels:
    email: EMAIL_ADDRESS

options.endpoint*

string | Pattern: ^https?://[^?#]+$

Base URL of the Presidio analyzer service. The analyze path is appended to it. A query string or fragment is not allowed.

options.credentials

object

Credentials presented to the analyzer.

options.credentials.authorization

string

Value sent as the Authorization header on each request.

options.language

string | Default: "en"

Language of the text, as understood by the analyzer.

options.threshold

number | Minimum: 0 | Maximum: 1 | Default: 0

Minimum analyzer confidence score, between 0 and 1, for an entity to count as detected. It is sent to the analyzer as score_threshold. When 0, the analyzer's own threshold applies.

options.timeout

string | Format: duration | Default: "PT5S"

Maximum time to wait for the analyzer, as an ISO-8601 duration. PT0S disables the timeout.

options.labels

object as map of string

Named labels to detect, each mapped to a Presidio entity type, such as EMAIL_ADDRESS, PERSON, PHONE_NUMBER, CREDIT_CARD or US_SSN. An entity type is either one of Presidio's built-in entities or a custom entity name of upper-case letters, digits and underscores, starting with a letter, such as EMPLOYEE_BADGE, for use with a custom recognizer registered on the analyzer.

labels:
  email: EMAIL_ADDRESS
  badge: EMPLOYEE_BADGE

When omitted, a label name used by a consumer is sent to the analyzer as the entity type itself. When present, at least one label is required, and only the labels listed can be referred to.