Gå til hovedinnhold

Redact tool

The Redact tool replaces sensitive spans in the source with protected placeholders before the content is processed (for example, before it is sent to a translation provider), and stashes the originals locally so a later unredact can restore them. It never emits the original value into the rewritten content. It is a recoverable transformer: it produces an edit plan — the span replacements plus the originals to vault — and the framework applier rewrites the source, rebasing surviving run-anchored overlays so an upstream annotator's terms or entities follow the rewrite.

Two detection backends can run, alone or together: rules matches literal terms and regular expressions from a rules file or inline rules, and entities redacts entity annotations already on the block (as produced by ai-entity-extract). When no detector is configured, rules is used. The originals are stored on the block as an in-process secret annotation by default, or written to a sidecar vault when a vault path is supplied (the external mode used by extractmerge).

IDredact
SourceBuilt-in
Categorytext-processing
Cardinalitymonolingual
Tagssecurity, redaction

Parameters

ParameterTypeDefaultDescription
detectorsstring[]Detection backends to run: rules and/or entities
entityTypesstring[]Entity categories to redact: person
placeholderstringVisible stand-in template; supports {category} and {n}
rulesPathstringPath to a redaction rules YAML file

Configure these parameters interactively and copy the flow-step YAML on the Tool Reference.

Examples

Redact by rules file

Replace spans matched by a rules file with the default placeholder.

detectors: [rules]
rulesPath: redaction-rules.yaml

Redact extracted entities

Redact person and organization entities already annotated on the block.

detectors: [entities]
entityTypes: [person, organization]

Processing notes

  • A recoverable transformer — it returns an edit plan and the framework applier rewrites the source, rebasing surviving overlays across the rewrite. The placement pass requires it to run before any step that sends source to a remote sink.

  • Originals are recorded as an in-process secret annotation by default, or in a sidecar vault when a vault path is supplied (external mode); the applier vaults them atomically with the rewrite.

  • A later unredact reads the recovery record to restore the originals.

Limitations

  • The entities detector only acts on entity annotations already present on the block; it does not detect entities itself — pair it with ai-entity-extract.

  • Inline rules and the sidecar vault path are supplied programmatically by callers (e.g. extract → merge) rather than through the schema fields documented here.

  • Operates on translatable source spans; dates, times, currencies, and measurements are excluded from the default entity set.

← Back to the Tool Reference