Glossary
Guardrail
A guardrail is a written-ahead limit that blocks an output or action outside policy. Unlike a prompt suggestion, a guardrail must be deterministic.
Also called گاردریل
A practical example: not publishing a figure that policy rejected, or blocking a customer message without approval. Those must not depend on the model behaving well.
Guardrails complement the autonomy ladder: they let the rung rise without removing the rail.
Let’s make your processes agentic
Where do we start?