# Guardrails

Define response boundaries and check the agent’s behavior.

## Set response boundaries

QwryAI Guardrails are instructions that describe response boundaries. Open Build → Instructions and edit Guardrails alongside the general instructions.

## Write clear boundaries

State how the agent handles missing facts, private information, unsupported promises, and requests outside its role. Keep these rules consistent with the actions and escalation paths you have actually enabled.

## Check the boundaries

1. Save the instructions.
2. Ask an ordinary supported question to confirm the rules do not prevent useful answers.
3. Ask for an unknown release date or unavailable account information.
4. Confirm the agent explains the limit and gives an appropriate next step.

## Guardrail scope

Instruction guardrails guide AI behavior. Allowed websites (Deploy → Specific domains), provider permissions, and action approval policies are separate controls. The rate limit, spam detection, and blocked countries are in agent Settings → General.

- [Instructions](/docs/user-guides/agent-management/instructions)
- [Action approvals](/docs/user-guides/agent-management/approvals)
- [Agent settings](/docs/user-guides/agent-management/settings)
