Configure classifier and guard experts

A published catalog names a classifier expert and an optional guard expert. This page describes how to enable these expert beans and set their properties.

Enable every expert referenced by the published catalog explicitly. The classifier remains expert.bean; an optional Boolean guard uses guard.expert.bean, guard.operation, and guard.rejectWhen in the catalog manifest. Missing guard beans fail startup before readiness or forward registration.

Repeat --expert for the classifier and guard:

--expert supportExpert=org.apache.camel.component.typesafeai.TypeSafeAiSemanticAdapter \
--expert injectionGuard=org.apache.camel.component.wolfdefender.WolfDefenderSemanticAdapter \
-p 'wsr.expert.injectionGuard.properties.modelDirectory={{env:WOLF_MODEL_DIRECTORY}}'

Bean properties use wsr.expert.<bean>.properties.<property> in deployment files or --property options. Camel resolves property placeholders and converts values to setter types before native evaluation validation and service startup. Unknown properties or invalid values fail startup with sanitized diagnostics. Credentials and model paths belong to the deployment; Barn catalogs contain no such settings.

The distribution includes camel-wolf-defender at the same Camel version as the semantic language. Provision its pinned model and tokenizer files locally, then set WOLF_MODEL_DIRECTORY to that directory. WSR never downloads model files. Camel owns each expert's lifecycle, including shutdown of model resources. Deploy this runtime before publishing guarded catalogs; restart for new revisions.

Next steps

• Run a published semantic route: start the runtime with the experts.
• Test Wanaku Semantic Router: run the Wolf Defender smoke test.