Configuring Agent Behavior Step

Field

Required

Description

Tips

The first words the AI Agent says when a call connects. Supports dynamic variables through + Add Variable.

💡 Keep the greeting concise and easy to understand.

💡 Include the AI Agent's name and job title to set clear expectations for the conversation.

💡 Tailor the greeting based on the call type.

Inbound Calls: Focus on welcoming the caller and offering assistance.

Outbound Calls: Clearly state the purpose of the call.

Example:

"Hi, I'm Adam from Maqsam, calling with an update regarding our latest product launch. Is this something you'd like to discuss?"

When enabled (ON), the AI Agent immediately stops speaking when interrupted by the caller.

When disabled (OFF), the AI Agent finishes its current response before allowing the caller to speak.

💡 Enable this option for inbound scenarios to create a more natural conversation experience, as the agent will pause once the caller begins speaking. 

💡 Consider disabling it for outbound notifications, reminders, or status updates where delivering the complete message is important.

Allows the AI Agent to end the call on specific conditions, e.g., keyword is detected, or the caller indicates they want to end the conversation.

⚠️ If disabled, the AI Agent will not attempt to end the call and the caller must end it manually.

Click "Configure Settings" to set conditions.

💡 Enable this action for all AI Agents to prevent calls from continuing unnecessarily.

💡 You can leave the configuration empty and use the platform's built-in call-ending flow, or configure a custom trigger and message if needed.

Select how the AI Agent should handle escalation requests.

Choose one of the following options:

• Escalate to a Human Agent: Transfers the caller to a live agent.

• Escalate by Sharing Contact Methods: Shares available contact information such as a phone number, WhatsApp number, email address, or website.

💡 Use Escalate to a Human Agent when live agents are available to receive transferred calls.

💡 Use Escalate by Sharing Contact Methods when you want callers to contact your team through alternative channels instead of transferring the call. 

💡 If your business does not offer live-agent transfers, you can still select Escalate by Sharing Contact Methods even if no contact details are configured. In this case, the AI Agent can acknowledge the request without attempting a transfer.

Contact methods are configured during the Company Information step.

Defines when the AI Agent should transfer the call to a human agent.

This can occur when the conversation reaches a specific point, when a customer mentions a specific keyword, or when the customer expresses interest in speaking with a human agent.

The message the AI Agent speaks before transferring the call.

Example: 

"I’m connecting you to a human agent now.”

The message the AI Agent speaks when a transfer cannot be completed because the call is outside working hours.

Example:

“I’m not able to transfer to a human agent as we are offline now. Would you like to leave your inquiry, and someone will contact you during our working hours ?”

The message the AI Agent speaks before sharing the configured contact method(s).

Example:

"Sorry, our team is currently unavailable. Would you like me to share our WhatsApp number so you can contact us directly?"

Text-based rules or restrictions that guide the AI Agent's behavior during calls. Supports dynamic variables through + Add Variable (e.g., never disclose internal pricing, always confirm appointment details before ending the call, etc.).

💡 Use clear and direct language.

💡 Use strong keywords such as MUST, SHOULD, and MUST NOT. 

💡Writing these keywords in CAPITAL LETTERS is recommended for emphasis. 

💡Add each guardrail on a separate line to improve readability and consistency. 

Free-text guidance that tells the AI Agent how to handle conversations and specific scenarios. Supports dynamic variables through + Add Variable.

⚠️ If left empty, the AI Agent will answer directly from the Knowledge Base.

💡 Define how the AI Agent should handle specific scenarios. 

💡 Include a brief description of your Knowledge Base structure (e.g., Pricing, Account Setup, Product Information). 

💡 Use CAPITAL LETTERS to emphasize important instructions. 

💡 Organize instructions in order of importance. 

💡 Use a hierarchical structure for complex instructions. 

# level 1
## level 2
### level 3
- Level 4
- - Level 5 
And so on

💡 Use numbered steps when defining sequential call flows. 

For example, an outbound call about upselling a specific service, the steps can be as follows

Step 1: After the greeting, ask customers if they would like to continue the call. If yes, go to step 2; if no, go to step 3

Step 2: Tell them the benefits of upselling and ask whether they are interested. Then go to step 4.

Step 3: Thank the customer on his/her time and end the call.

Step 4: ... and so on.

💡 You can use an LLM to assist with drafting the instructions for this section by providing context that this is for a voice AI agent and the requested call flow. Ensure you use Gemini on your Maqsam account to protect the client’s data.

🟥 Guardrails 

🟨  Instructions 

 We already added generic guardrails for all AI Agents generated at Maqsam (e.g. agents are always respectful and avoid negative language) 

 Can include the knowledge base directly if it is small and simple, which helps the AI respond faster 

 Used to add additional business-specific guardrails on top of the default system behavior 

 Used when you want the agent to handle a clear scenario in a structured way 

 Focus on rules and restrictions for agent behavior 

 Focus on scenario-based behavior and call flow logic 

 Examples:

MUST NOT disclose pricing 

MUST ALWAYS verify identity before proceeding

MUST NOT transfer the call unless a valid trigger is met

Examples: 

If you want the agents to collect customer information whenever they need to ask about a specific service, you can add a prompt in the instructions as follows: "When customers ask about the loan application, collect their name and phone number, then tell them that the team will contact them soon."

Outbound calls with a specific sequential flow, for example: "After the greeting, ask the client if they are interested in continuing the call; if yes, respond with this answer, etc."