Skip to content

Thinking mode, guardrails, and evals in the builder

The agent builder gained advanced controls: turn on extended thinking for harder reasoning, set input and output guardrails to enforce your rules, and attach evaluators that score responses automatically.

The agent builder gained advanced controls: turn on extended thinking for harder reasoning, set input and output guardrails to enforce your rules, and attach evaluators that score responses automatically.

  • Thinking mode for deeper reasoning
  • Input guardrails to block off-topic or unsafe prompts
  • Output guardrails to validate responses before they’re delivered
  • Attachable scorers for automatic quality checks