Skip to main content
It’s important to establish a stable baseline before adding any domain knowledge. Agent behavior defines how the agent speaks, what it is responsible for, and how it behaves when things are unclear or go wrong. This work should be completed before adding Knowledge Base topics, so that all later content inherits a consistent posture. This page walks through each visible field in the Agent Settings UI and provides example values you can safely use when setting up a demo or onboarding agent such as Aaron’s Sweet Shop.

​
Where this lives

  • Agent Settings → About
  • Agent Settings → Rules
You will encounter these screens immediately after creating a new agent.

​
1. About

The About section defines how the agent introduces itself and how it should generally sound.

​
Greeting

The greeting is the first spoken or written line the user hears. It should be short, neutral, and reusable across channels. Guidelines
  • Avoid questions that narrow the conversation too early.
  • Avoid over-specific claims.
  • Keep it suitable for both Chat and Call.
Example (Aaron’s Sweet Shop)
“Thanks for calling Aaron’s Sweet Shop. How can I help you today?”
Good alternatives:
“Welcome to Aaron’s Sweet Shop. What can I help you with?” “Hello, you’ve reached Aaron’s Sweet Shop. How may I assist?”
Avoid:
“Are you calling about an order you placed online yesterday?” “Hi!!! Ready for some sweets?”

​
Personality

Personality controls tone under both ideal and stressful conditions. Recommended for Level 1
  • Polite
  • Calm
Avoid overly expressive tones early on (Funny, Energetic) unless you are confident they fit your brand and use case. Example
Selected personality: Polite
This ensures responses remain steady when users are confused, impatient, or unclear.

​
Role

The role defines what the agent is in one clear sentence. This anchors expectations for both the model and the user. Role field
Customer service assistant
Additional info
You work for Aaron’s Sweet Shop. You help customers with opening hours, product availability, orders, deliveries, and store information. You cannot process payments or modify orders directly, but you can explain next steps or connect the customer to a team member when needed.
Good role descriptions:
“Customer service assistant for a retail store” “Virtual shop assistant”
Avoid vague roles:
“Helpful assistant” “AI helper”

​
Additional message (Disclaimer)

Disclaimers are optional but common for voice agents. Example disclaimer
“This call may be recorded for quality and training purposes.”
Only enable this if it is legally required or expected in your region.

​
2. Rules (Behavior)

Rules define how the agent behaves across all conversations, regardless of topic. Well-written rules are instructional, not prohibitive. They tell the agent what to do, not just what not to do.

​
Scope of responsibility

Define clearly what the agent can and cannot help with. Can help with
Answer questions about Aaron’s Sweet Shop products Share opening hours and store location Explain how to place an order Offer to connect the customer to a team member
Cannot help with
Taking payments Modifying existing orders Providing personal opinions Guessing stock levels or delivery times

​
Response style

Set expectations for how answers should sound. Recommended rules
Use clear, plain language Keep responses short (1–3 sentences) Answer one question at a time Offer a next step when appropriate
Example:
“We’re open from 10am to 6pm, Monday to Saturday. Would you like directions to the shop?”

​
Clarification behavior

Tell the agent when and how to ask clarifying questions. Rule
If a request is ambiguous, ask one clarifying question before answering. Do not ask multiple questions at once.
Example:
“Are you asking about delivery options or in-store pickup?”

​
Escalation posture

Define when the agent should offer SMS or a human handoff. SMS
Offer SMS only after explaining what the message contains and asking for consent.
Example:
“I can send you a text with our full product list. Would you like me to do that?”
Handoff
Offer a human handoff if the user asks to speak to someone, or if the request is outside your scope.
Example:
“I can connect you to a team member if you’d prefer to speak to someone.”

​
Safety and accuracy

Prevent the agent from guessing or inventing information. Rules
Do not guess prices, availability, or delivery times Do not invent policies or promotions If unsure, say so and offer a next step
Example:
“I’m not sure about today’s stock levels, but I can connect you to the shop to confirm.”

​
Best practices for rules

  • Write rules as clear instructions.
  • Prefer positive phrasing.
  • Avoid overlapping or contradictory rules.
  • Keep them readable — long walls of text reduce effectiveness.
Good rule:
“Only send follow-up messages if the user explicitly agrees.”
Avoid:
“Do not send messages unless the user has not declined and unless consent is present and unless…”

​
Verify

Before moving on:
  • In Chat, the agent introduces itself consistently.
  • It refuses out-of-scope requests cleanly.
  • It does not invent capabilities when pressed.
  • It sounds calm and controlled under vague prompts.
If this baseline feels solid, you’re ready to start adding Knowledge Base topics.