Setting Boundaries — The Rules Node (Red)

Tell your agent what it must never do. This is your safety net.

5 min read

What this does

Rules tell your agent what it must NEVER do. Without rules, your agent might make up information it doesn't actually know, reveal private details about your business, or wander completely off-topic. Rules are your safety net — they keep the agent inside the lines.

Think of it this way: every other node tells the agent what TO do. This one tells it what NOT to do. Both are equally important.

How to open the Rules node

Click the red box labeled "Rules" on your prompt map. It will expand to show all the settings inside.

Pre-built rules — your starting checklist

When you open the Rules node, you'll see three groups of pre-built rules already there. Most are checked by default. We recommend leaving them checked unless you have a very specific reason to turn one off.

Safety Rules

These prevent your agent from generating harmful, offensive, or inappropriate content. They also stop it from making up facts or pretending to be a real person. Without these, your agent could say something that damages trust or even creates legal issues. Leave these on.

Brand Protection Rules

These keep your agent on-brand. It won't badmouth competitors, make promises you can't keep, or say things that contradict your messaging. Without these, your agent might agree with a customer who says your product is bad, or promise a feature you don't offer. That's a headache you don't want.

Conversation Quality Rules

These keep responses helpful and on-topic. The agent won't ramble, go on tangents, or answer questions it wasn't asked. Without these, you might get responses that are technically accurate but painfully long, or answers to questions nobody asked. These rules keep things tight and useful.

Think of the pre-built rules like training wheels. They handle the obvious stuff so you can focus on rules that are specific to YOUR business.

Custom rules — make it yours

You can add up to 5 custom rules. These are where you get specific about what your agent should never do based on your particular business. Here are some real examples across different industries:

Bakery

"Never claim products are allergen-free — direct allergy questions to call us."

E-commerce Store

"Never offer discounts above 10% without manager approval."

SaaS Company

"Never share product roadmap or upcoming features."

Real Estate Agency

"Never give specific property valuations."

Restaurant

"Never take reservations — direct to OpenTable link."

Consulting Firm

"Never give specific advice — recommend scheduling a paid consultation."

Notice the pattern? Each rule protects the business from something that could go wrong if the agent answered on its own. Think about what questions your customers ask where the wrong answer could cause a real problem. Those are your custom rules.

Click "Add Custom Rule" inside the Rules node.

Type your rule in plain English. Be specific — "Don't talk about pricing" is better than "Be careful."

Repeat for up to 5 rules. You don't need to use all 5. Two or three great rules are better than five vague ones.

Write rules the same way you'd tell a new employee on their first day. "Never promise same-day delivery" is clear. "Be mindful of delivery expectations" is not.

Escalation settings — when to call a human

Some conversations shouldn't stay with the AI. Escalation settings let you define exactly when your agent should stop trying to help and hand things off to a real person.

There are two parts to set up:

Triggers — what topics should hand off

These are the situations where your agent should stop and get help. Common examples: billing disputes, refund requests over $100, legal questions, harassment or threats, complaints that need a manager, and anything involving sensitive personal information.

Add each trigger as its own item. The more specific you are, the better the agent knows when to step aside.

Action — what the agent says when it escalates

When a trigger fires, what should the agent actually say to the customer? Walk through each option and pick the one that fits your setup. You might have it say something like "I want to make sure you get the best help on this — let me connect you with our team" and then provide a link, email, or phone number. The key is to make the handoff feel smooth, not like the agent just gave up.

Don't skip escalation settings. An agent that tries to handle everything — including things it shouldn't — is worse than one that knows when to tap out. Customers respect "let me get you to someone who can help" much more than a wrong answer.

Opening the Rules node → scrolling through pre-built rule groups → adding a custom rule → setting up an escalation trigger and action message

Video • 45 seconds

You just set the boundaries for your agent — what it must never do and when to ask a human for help. These rules are your safety net, and they keep your agent from making mistakes that could hurt your business or frustrate your customers. Next up: the most important step of all.

📚 Knowledge Node

💬 Examples Node