Guardrails
Ensure safe and constrained LLM output.
Guardrails constrain what agents may say or do — topic blocks, policy rules, and safe fallbacks applied at runtime.
Configure guardrails
Open Build → Agents and edit an agent in the Configuration workspace.
Go to the Guardrails tab and add rules that match your compliance needs.
Test edge cases in Agent Console — try prompts you expect to be refused or redirected.
Deploy or save changes so live channels pick up the updated policy.
Guardrails complement your system prompt on the Brain tab; they enforce hard limits even when the model drifts.