How to Set Up Instructions and Guardrails for Your AI Agent?

Instructions tell your AI agent how to reply; guardrails define what it must never do. Both live as toggleable rule cards in the Behavior section of your agent.

What Is AI Agent?

Instructions and Guardrails are two tabs in your agent’s Behavior section. Instructions are positive rules — how to greet, how long to answer, when to ask clarifying questions — while guardrails are hard limits, such as never sharing sensitive personal data or inventing prices. Each rule is a card with a toggle, and you can write your own, generate them from your trained knowledge, or import them from a document.

Why Use It?

Rules turn a generic chatbot into an agent that sounds like your team and stays inside your policies.

• The agent’s replies don’t sound like your brand — instructions set the greeting, tone-matching, and reply length so every answer fits your voice.
• You worry the AI will promise something it shouldn’t — guardrails block invented prices, legal or medical commitments, and sensitive data handling.
• Writing rules from scratch is slow — Generate from knowledge drafts rules from your trained sources in seconds, ready to edit.

Example Use Cases

• A support agent adds the instruction to ask one clarifying question when a request is ambiguous, cutting back-and-forth.
• A team manager switches on the guardrail that escalates angry or churn-risk language straight to a human.
• A small business owner imports the company policy document so the agent’s rules match what staff already follow.

Step-by-Step Instructions

Step 1: Open your agent, go to Behavior, and select the Instructions tab.

Step 2: Switch each instruction card on or off with its toggle to control which rules the agent follows.

Step 3: Click + Add instruction to write a custom rule in your own words.

Step 4: Click Generate from knowledge to let the agent draft instructions from your trained sources, then edit or toggle the results.

Step 6: Click + Add guardrail or Generate from knowledge to add safety rules specific to your business.

Step 7: Open Preview, send a message that tests a rule, and check the Agent Trace panel to confirm which instruction or guardrail fired.

Pro Tips & Interactive Settings