Call Configuration

The Call Configuration tab controls how your agent behaves during voice calls. Here you define greeting strategies, ambient sounds, timeout behaviors, and powerful voice tools that give your agent capabilities like transferring calls and collecting keypad input.

---

Conversation Settings

Greeting Message

Configure how your agent greets users at the start of a voice call. In voice calls, the agent always initiates the conversation—the greeting strategy controls what the agent says, not whether it speaks first.

Greeting Strategies

Strategy	Description
AI Generated	Agent creates a dynamic greeting based on context and instructions every time
Smart (Contextual)	Uses AI-generated greeting if conversation history exists, otherwise falls back to the fixed message
Fixed Message	Agent always says the exact message you specify—consistent and predictable

Agent Always Speaks First

In voice calls, the agent always initiates the conversation. There is no "wait for user" option. The greeting strategy only controls how the greeting content is generated.

When to Use Each Strategy

• AI Generated: Best when you want natural, varied greetings that adapt to context. The agent uses its instructions to craft appropriate openings.
• Smart (Contextual): Ideal for returning users—generates a personalized greeting when there's conversation history, but uses your fixed message for first-time callers.
• Fixed Message: Choose this when consistency matters, such as compliance scenarios or when you need exact wording every time.

Example fixed greeting:

"Hello, thank you for calling TechCorp support. My name is Sarah, how can I help you today?"

Using Variables

If you're using dynamic variables like , consider AI Generated or Smart strategies so the agent can incorporate personalized elements naturally.

---

Background Sounds

Add audio to make calls feel more natural and reduce perceived latency.

Thinking Sound

Audio that plays while the agent is processing a response. This fills the silence that would otherwise occur during AI inference time.

The platform provides predefined thinking sounds that you can enable or disable:

• Enable/Disable: Toggle whether thinking sounds play during processing
• Volume: Adjust the audio level (0-100%)

Why Use Thinking Sounds?

A 2-second silence feels like an eternity on a phone call. Background sounds significantly improve perceived responsiveness—the same delay feels like the agent is "working on it" rather than frozen.

Ambient Sound

Continuous background audio throughout the call. Choose from predefined ambient sounds to create a more natural atmosphere.

• Enable/Disable: Toggle ambient sounds on or off
• Volume: Adjust the audio level (0-100%)

Available Sounds

The platform offers several predefined sounds for both thinking and ambient audio. These are optimized for voice calls and designed to be unobtrusive while filling silence naturally.

---

Duration Limits

Maximum Call Duration

Set the maximum length for any single call (0.5 to 60 minutes). When this limit is reached, the agent will gracefully end the conversation.

Configuration:

• Time in minutes: Set the maximum duration (0.5-60 minutes)
• Message (optional): Custom message the agent says when ending due to time limit

Typical settings:

• Customer support: 15-30 minutes
• Sales qualification: 10-15 minutes
• Simple inquiries: 5-10 minutes

Inactivity Timeout

Configure how the agent handles unresponsive users. The system can prompt users to check if they're still there before ending the call.

Configuration:

• Enable/Disable: Toggle inactivity detection
• Time in seconds: How long to wait before first prompt (1-300 seconds)
• Presence attempts: Number of "Are you still there?" prompts (0-5)
• Presence message: The message used to check if user is present

How Timeout Works

The total timeout is calculated as: time_in_seconds + (time_in_seconds × presence_attempts)

Example: With 30-second timeout and 2 presence attempts:

• After 30s of silence → Agent says presence message
• After another 30s → Agent says presence message again
• After another 30s → Agent ends the call

Total time before ending: 90 seconds

Timeout	Presence Attempts	Total Time	Use Case
30 seconds	1	60 seconds	Standard conversations
30 seconds	2	90 seconds	Complex topics
60 seconds	2	180 seconds	Users may step away briefly

Customize the Presence Message

The default presence message works well, but you can customize it to match your agent's personality:

• Formal: "I haven't heard from you. Are you still on the line?"
• Friendly: "Hello? Take your time if you need to!"
• Direct: "Are you still there?"

---

Voice Tools

Voice tools give your agent superpowers during phone conversations. These capabilities allow the agent to take real actions beyond just talking.

Overview

Tool	What it Does	Availability
End Dialog	Agent can hang up when conversation is done	All channels
Transfer Call	Transfer to human agent or another extension	SIP only
Send DTMF	Navigate automated phone menus (IVR)	SIP only
Receive DTMF	Collect data through phone keypad (e.g., PIN)	SIP only (Beta)

SIP-Only Tools

Tools marked SIP only work exclusively during SIP/WebRTC calls. They won't work in Voice Playground or WhatsApp voice messages.

---

End Dialog

Allows your agent to gracefully end the call when the conversation is complete. The agent will say goodbye politely before hanging up.

When to use:

• Customer's request has been resolved
• Customer asks to end the call
• Conversation has reached a natural conclusion
• No further assistance is needed

Configuration:

1. Enable End Dialog
2. (Optional) Customize the description to guide when the agent should end calls

Collecting Data When Calls End

Use the JSON Schema field to gather structured information when the call ends. This data appears in webhooks and conversation history.

{
  "type": "object",
  "properties": {
    "reason": {
      "type": "string",
      "enum": ["resolved", "transferred", "callback_requested", "customer_ended"],
      "description": "Why the call ended"
    },
    "satisfaction": {
      "type": "integer",
      "minimum": 1,
      "maximum": 5,
      "description": "Customer satisfaction if mentioned"
    }
  }
}

---

Transfer Call

Lets your agent transfer the call to another destination—a human agent, a specific department, or an external number.

SIP Only

This tool only works during SIP calls. It won't function in Voice Playground or WhatsApp.

When to use:

• Customer asks to speak with a human
• Issue requires specialized support
• Business rules require escalation
• Customer needs a specific department

Configuration:

1. Enable Transfer Call
2. Add your transfer destinations (SIP URIs or phone numbers)
3. Write a description explaining when to transfer to each destination

Transfer Destination Formats

Destination	Format	Example
SIP Address	sip:name@domain	sip:support@company.com
Phone Number	sip:+number@provider	sip:+15551234567@sip.provider.com
Extension	sip:ext@pbx	sip:100@pbx.internal

Real-World Scenario: Call Center Escalation

Situation: Customer calls about a technical issue beyond AI capabilities.

Agent: "I understand you're experiencing issues with your enterprise license. Let me transfer you to our technical support team who can help you directly. Please hold for a moment."

Agent transfers to sip:tech-support@company.com

Write Clear Transfer Instructions

In your agent's instructions, be specific about when to transfer:

• "Transfer to technical support for server configuration or enterprise features"
• "Transfer to billing for refund requests or payment disputes"
• "Transfer to sales when customers ask about upgrading their plan"

---

Send DTMF

Allows your agent to send touch-tone signals (the beeps when pressing phone keys) to navigate automated phone menus.

SIP Only

This tool only works during SIP calls.

When to use:

• Agent needs to navigate an external phone system (IVR)
• Opening doors or gates controlled by phone systems
• Interacting with automated services requiring keypad input

Available tones: 0-9, *, #

Real-World Scenario: Smart Building Entry

Situation: Your agent receives calls from building intercoms and can buzz visitors in.

Visitor (via intercom): "Hi, I'm here for the 2pm meeting with Sarah."

Agent: "Welcome! Let me verify... Yes, I see you're expected. I'm opening the door now."

Agent sends DTMF "9" to unlock the door

Agent: "The door is open. Please proceed to the 3rd floor. Have a great meeting!"

Real-World Scenario: IVR Navigation

Situation: Your agent calls an external service and navigates their phone menu.

External IVR: "Press 1 for English, 2 for Spanish" Agent sends "1"

External IVR: "Press 1 for account information, 2 for support" Agent sends "2"

Agent is now connected to support

---

Receive DTMF Beta

Allows your agent to collect information through the phone keypad. Perfect for sensitive data like PINs or account numbers that customers may prefer to type rather than speak.

Beta Feature & SIP Only

This feature is in beta. It only works during SIP calls and cannot be tested in Voice Playground.

When to use:

• Collecting sensitive data (SSN, PIN, account numbers)
• Data that's easier to type than speak
• Situations requiring precise numeric input
• Compliance requirements mandating keypad entry

Configuration:

1. Enable Receive DTMF
2. Set the Default Timeout (1-60 seconds)
3. (Optional) Customize the description

How It Works

When enabled, your agent can ask for keypad input. The agent automatically determines how many digits to collect, which key ends input, etc., based on conversation context. You only configure the timeout.

Real-World Scenario: Identity Verification

Situation: A bank's AI agent verifies customer identity before discussing account details.

Agent: "For security purposes, I need to verify your identity. Please enter your 9-digit Social Security Number using your phone keypad. Press pound when done, or star to start over."

Customer enters: 1-2-3-4-5-6-7-8-9-#

Agent: "Thank you. I received: one, two, three, four, five, six, seven, eight, nine. Is that correct?"

Customer: "Yes, that's right."

Agent: "Perfect, I've verified your identity. How can I help you today?"

Critical: Always Confirm Input

Never Skip Confirmation

Always have your agent read back digits and ask for confirmation. DTMF input can have errors (wrong key, missed digit), and processing incorrect data causes serious problems.

Add this pattern to your agent instructions:

When collecting keypad input:
1. Tell the customer what to enter (mention * to restart, # to finish)
2. Collect the digits
3. Read them back one by one: "I received: one, two, three..."
4. Ask: "Is that correct?"
5. If they say no, offer to collect again
6. Only proceed with confirmed data

---

Voice Tools Best Practices

1. Always Confirm DTMF Input

Read back each digit individually ("one, two, three" not "one twenty-three") and wait for clear confirmation.

2. Offer Alternatives

Not everyone can easily use a phone keypad. Always offer the option to speak information instead, especially for non-sensitive data.

3. Set Appropriate Timeouts

Scenario	Recommended Timeout
Data customer knows by heart	10-15 seconds
Data they might need to look up	25-30 seconds

4. Provide Clear Instructions

Before collecting DTMF, tell customers:

• Exactly what to enter
• How many digits
• What key finishes entry (#)
• What key restarts (*)

5. Test with Real Calls

Voice Playground Limitation

SIP voice tools don't work in Voice Playground. You must test with actual SIP calls before deploying to production.

---

Related Documentation

• Voice Settings - Configure voice and TTS
• Instructions - Write effective agent prompts
• SIP Deployment - Set up SIP for voice calls
• Practical Guide: SIP Voice Tools - Detailed tutorials