What is Guardrails?

Constraints on what an agent may say or do.

Guardrails are the rules that bound an agent's behaviour: topics it must decline, claims it must not make, actions it may not take without confirmation. They are enforced around the model rather than trusted to it.

This is an industry glossary. A term defined here is not a statement that Drop AI provides it — see the pricing page for what is included.

System prompt

The written instructions that define an agent's behaviour.

Hallucination

A confident answer that is not true.

Prompt injection

Input crafted to override an agent's instructions.

Voice AI

More in this category

Voice agent Conversational AI ASR STT TTS LLM System prompt Function calling Tool use Knowledge base RAG Grounding

Transform every customer conversation.

Start free, or talk to us about what a custom rollout looks like for your team.

Try for Free Contact Sales