What is Guardrails?
Guardrails are the rules and checks around a model's behaviour — what it refuses, what it must include, and validators that catch bad output before it reaches a user.
Guardrails live in three places: in the prompt (hard rules and refusals), in code (validators that check format, banned content or missing fields), and in process (a human reviews anything high-stakes). The prompt alone is not a guardrail — models can be talked out of instructions.
For personal use, a guardrail is simply a rule you never drop: "cite the passage", "flag uncertainty", "don't write the essay for me". For products, it's the difference between a demo and something you can ship.
- State the rule, the exact behaviour when it's triggered, and the fallback.
- Validate structured output in code; don't trust the model to self-report.
- Keep humans in the loop for money, health, legal and safety.
Use it right now
Ask our brain anything on the homepage — it remembers the whole conversation — or write a brief in the Studio and see the prompt it compiles to.