Rylvo
All updates
New feature

Typed decisions with TypeSafe Jev

Guardrails, routing, knowledge relevance, and other short decisions can use TypeSafe Jev, with trace visibility and an automatic fallback.

JevGuardrailsKnowledge BaseObservability

Rylvo now uses TypeSafe Jev for short, typed decisions in several agent workflows. It chooses from defined options or scores a signal; the model that writes your agent's reply remains separate.

Where it helps

  • Custom guardrail evaluations for the same text can be checked together.
  • Workflow stage and sentiment, intent-only group routing, flow intent branches, and fan-out vote selection can use typed decisions.
  • Knowledge Base Self-RAG can grade retrieved passages together.
  • Mission Control can add model-scored frustration, urgency, and cancellation signals to live conversations.

Jev decisions appear in traces with their answers, probabilities, latency, tokens, and cost. When Jev is unavailable, the existing decision path takes over automatically. Organizations can turn Jev off in Settings → Security. See the documentation and subprocessor list for details.