classify Model

Defines a model that rejects a message value when one or more classifiers detect a configured label in it.

The full message value is buffered and decoded as UTF-8 text before being classified. The value itself is never modified, it is only accepted or rejected. The value is accepted only when every referenced classifier reports that none of its listed labels are present, and rejected as soon as any one does. A value larger than 1 MiB, or one that a classifier fails to classify, is rejected.

model: classify
reject:
  secrets0:
    - secret.aws_key
    - secret.github_token
  moderator0:
    - prompt.injection

Configuration (* required)

model: classify

const

Specifies the model is classify.

reject*

object as map of named array properties

Labels to reject, grouped by the name of the configured classifier that detects them. Each key is the name of a classifier (patterns, presidio, or semantic) and each value is a non-empty list of that classifier's labels. At least one classifier is required.