Guardrails

What are guardrails?

Guardrails are security rules and limits that restrict what artificial intelligence can do. They act as safety rails that keep AI within the parameters defined by your organization, ensuring appropriate responses and preventing unwanted behavior.

Types of guardrails

• Token limits — Controls the maximum length of AI-generated responses.
• Content restrictions — Defines topics or response types that the AI should not generate.
• Tool access — Controls which tools an AI agent can use (query data, send emails, etc.).
• Cost limits — Sets a credit cap that the AI can consume per execution.
• Output validation — Verifies that the AI response meets an expected format or structure.

Configure guardrails

Guardrails are configured at the node level in the workflow or at the agent level. When configuring an AI node or an agent, you will find the "Guardrails" section where you can define the applicable rules. Cost and token limits are managed from the Resource Settings section.