Content safety rules — configured from the dashboard.
Guardrails work today from the dashboard: configure blocked categories and prompt-injection detection at Settings → Guardrails. They apply account-wide to every request, including API-key requests — a blocked request fails with the guardrail_blocked error.
Not yet available. A key-authed guardrails API and a per-request
guardrailsfield are tracked to be built — policies are managed in the dashboard only for now.
guardrail_blocked error shape