Skip to main content

Overview

Agent Phone Integration allows you to connect your conversational AI agents to phone numbers, enabling customers to call and interact with your agents directly. This feature uses Twilio as the phone service provider to handle incoming calls and route them to your configured agents.

Prerequisites

Before setting up phone integration for your agents, ensure you have:
1

Twilio Phone Provider Integration

Twilio integration configured in your organization settings with active credentials by an organization admin
2

Active Agent

At least one agent created and configured in your app with proper training instructions
3

Twilio Phone Number (Optional)

A phone number provisioned in your Twilio account - you can also provision through the platform

Setting Up Phone Integration

1

Configure Phone Provider

Navigate to your Organization Settings and ensure you have a Twilio phone provider configured:
  1. Go to Organization SettingsIntegrationsPhone
  2. If no phone providers are configured, click Phone Provider to add Twilio credentials
  3. Verify your Twilio integration shows as active
Your Twilio integration must show as “Active” before proceeding to the next step.
2

Provision a Phone Number

Add a phone service from your app:
  1. Open your App in Elementum.
  2. Click App Details.
  3. In the Phone Services section, add a new service. Choose how the number is sourced:
Let Elementum provision a new number through your Twilio integration:
  1. Start adding a phone service and select provisioning a new number.
  2. Complete the service fields (see next step).
Platform provisioning handles Twilio configuration for the new number automatically.
3

Configure Phone Service

When you add or edit a phone service, set:
  • Number Type: New number (platform-provisioned) or existing Twilio number
  • Area Code: Geographic area code when provisioning a new number
  • Phone Provider: Your organization’s configured Twilio provider
  • Agent: The agent that handles calls on this service
  • Default Language: Primary language for calls on this service
  • Additional Supported Languages: Optional extra languages beyond the default. See Supported Languages. Enabling many languages may reduce performance.
Select the LLM for the voice agent. Current options include GPT-4o Mini and Grok 4 Fast Non-Reasoning. Models use platform-level configuration; custom credentials are not configured here.
When enabled, the platform generates an initial greeting the agent speaks when a call connects. You can also set a custom first message on the agent instead.
Click Save when finished.
4

Configure Agent for Phone Calls

Once your phone service is created, configure how your agent behaves during phone conversations:
  • Training Instructions: Specific guidance for phone interactions (e.g., “Keep responses concise for phone calls”)
  • First Message: Custom greeting message (if not using auto-generation)
  • Conversation Timeout: Set timeout duration for inactive calls
  • End Call Training: Instructions for how the agent should end conversations
See Best Practices for phone agent training guidance.

Configure App Phone Services Settings

Manage phone services in your app:
  1. Open your App in Elementum.
  2. Click App Details.
  3. In the Phone Services section, manage or configure a service using the following fields:
    • Number Type
    • Area Code
    • Phone Provider
    • Agent
    • Default Language
    • Additional Supported Languages
From Phone Services, you can see provisioned numbers, which agent is assigned, service status, and call activity or logs.
After these steps, open Automations in your app’s menu to add Phone Agent actions to your workflows.

Agent Tools for Phone Calls

You configure agent tools—including tools used on phone calls—while you build and configure the agent. The endConversation tool below is specific to phone interactions.

EndConversation Tool

Agents have access to a special endConversation tool specifically designed for phone interactions. Purpose:
  • Allows agents to proactively end phone calls when appropriate
  • Provides clean closure for support interactions
  • Prevents calls from remaining open unnecessarily
  • Ensures proper conversation lifecycle management
Critical Configuration: Train your agents on when and how to use the endConversation tool for optimal user experience.
When you have fully resolved the caller's question or completed the requested task, 
use the endConversation tool to end the call. Always provide a brief summary 
of what was accomplished and any next steps before ending the conversation.
Use the endConversation tool when:
  • Support issue has been fully resolved
  • Caller’s question has been completely answered
  • Required information has been collected
  • Follow-up actions have been clearly communicated
  • Caller confirms satisfaction with the resolution

Initiating Outbound Calls

Beyond receiving inbound calls, agents can proactively initiate outbound calls to customers as part of automated workflows. This enables your agents to make follow-up calls, appointment reminders, and other proactive outreach automatically.

Overview

The Initiate Call automation action allows workflows to trigger outbound phone calls from your AI agents. It uses the same agent and phone service configuration as inbound calls.
The action is fire-and-forget: the workflow continues immediately without waiting for the call to complete. Use the Agent Conversation Ended trigger in a separate automation to act on call outcomes (see Workflow Triggers).

How It Works

  1. Workflow Trigger: An automation workflow reaches an “Initiate Call” action
  2. Call Initiated: The system asynchronously starts a phone call to the specified number
  3. Agent Conducts Call: Your configured AI agent handles the conversation
  4. Workflow Continues: The workflow proceeds to the next action without waiting
  5. Call Outcome Captured: Use the Agent Conversation Ended trigger in another automation to process results

Configuring the Initiate Call Action

When adding an “Initiate Call” action to your workflow, you’ll configure:
Action Name
  • Descriptive name for the call action to identify it in your workflow
AI Agent
  • Select which agent will conduct the call
  • The agent must be properly trained for phone interactions
  • Agent will have access to any context you provide
Phone Service
  • Choose which phone service/number will be used to place the call
  • Must be a phone service you’ve already configured in your app
Phone Number to Call
  • Specify the recipient’s phone number
  • Can use:
    • Record Field: {{customer.phone_number}}
    • Previous Task Output: {{search_customer.phone}}
    • Static Value: +1-555-123-4567
Default Language
  • Set the primary language for the call
  • Agent will conduct conversation in this language
Related Record
  • Associate the call with a specific record
  • Provides context for the agent and enables better record keeping
  • Agent can access related record data during the call
Additional Context
  • Provide specific instructions or information for the agent
  • Examples:
    • “This is a follow-up call for support ticket #12345”
    • “Customer requested callback about billing issue”
    • “Appointment reminder for 2pm tomorrow”
  • Helps the agent conduct a more informed conversation

Example Workflows

TRIGGER: Support Ticket Created
  → IF priority = "High"
    → Initiate Call
      → Agent: Support Agent
      → Phone Service: Main Support Line
      → Phone Number: {{ticket.customer_phone}}
      → Related Record: {{ticket.id}}
      → Context: "Follow up on high-priority ticket {{ticket.number}}"
      → Language: English
  → Continue with other workflow actions

Fire-and-Forget Execution

You cannot reference call outcomes in the same workflow. Create a separate automation with the Agent Conversation Ended trigger to act on call results.
Example: processing call outcomes in a separate automation
TRIGGER: Agent Conversation Ended
  → IF conversation.type = "Voice"
    → Post Comment (on related record)
      → Comment: "Call completed: {{conversation.summary}}"
    → IF conversation contains "needs follow-up"
      → Create Record (Follow-up Task)

Best Practices for Outbound Calls

  • Agent Training: Train agents specifically for outbound scenarios with appropriate opening statements and call objectives
  • Context Matters: Always provide relevant context to help the agent conduct meaningful conversations
  • Timing Considerations: Consider time zones and appropriate calling hours in your workflow logic
  • Fallback Handling: Use Agent Conversation Ended to detect no-answers and create follow-up tasks
Agent Training Tips for Outbound Calls:
Example Outbound Agent Instructions
You are making an outbound call to the customer. Start by:
1. Introducing yourself and the company
2. Stating the purpose of your call clearly
3. Confirming you're speaking with the right person
4. Proceeding with the conversation objective

Keep the call focused and respect the customer's time. If they're unavailable,
offer to call back at a better time or provide alternative contact methods.

Use the endConversation tool when you've completed the call objective or if
the customer requests to end the call.

Monitoring Outbound Calls

Outbound calls create app records and transcripts like inbound calls. Use the Chat/Conversation interface on the call record to review transcripts and metadata. Use Agent Conversation Ended triggers to create follow-up tasks, send notifications, or generate analytics (see Integration with Workflows).

Common Use Cases

Automatically call customers before scheduled appointments to confirm attendance and reduce no-shows.
Proactively reach out to customers after support tickets are resolved to ensure satisfaction.
Notify customers about important order status changes when immediate communication is critical.
Contact customers with overdue payments with friendly reminders and payment options.
Conduct automated satisfaction surveys or gather feedback through phone conversations.
Alert customers about urgent issues, security concerns, or time-sensitive matters.

Supported Languages

Set Default Language on each phone service. Use Additional Supported Languages when you need more than one language on that service.
Enabling additional languages can reduce performance; enable only if your use case requires it.

Available Languages

  • Arabic
  • Chinese (Mandarin) — Simplified and Traditional
  • Chinese (Cantonese) — Traditional
  • Dutch
  • English
  • French
  • German
  • Hindi
  • Hungarian
  • Indonesian
  • Italian
  • Japanese
  • Korean
  • Portuguese
  • Russian
  • Spanish
  • Thai
  • Turkish
  • Vietnamese

Performance Considerations

When using multiple languages: test thoroughly, monitor call quality and response times, and prefer single-language mode when possible for best performance.

Testing Your Phone Integration

Testing Call Flow

  1. Make a Test Call: Dial your provisioned phone number
  2. Verify Connection: Ensure the call connects and the agent responds
  3. Test Agent Responses: Interact with the agent to verify it follows training instructions
  4. Check Call Logs: Review call logs in your app to ensure data is captured correctly

Troubleshooting Common Issues

Common Solutions:
  • Verify Twilio integration is active
  • Check that the phone number is properly registered
  • Ensure your Twilio account has sufficient credits
Common Solutions:
  • Verify the agent is properly configured and active
  • Check that the LLM model is available
  • Review training instructions for phone-specific guidance
Common Solutions:
  • Reduce Additional Supported Languages if not needed
  • Review agent training to ensure responses are concise
  • Check Twilio service status

Integration with Workflows

Automations can trigger outbound calls and process call outcomes. For outbound call setup, see Initiating Outbound Calls.

Call Data Capture

When a phone call is received or initiated, a record is automatically created in your app with call information:
  • Phone number (caller or recipient)
  • Call start time
  • Call duration
  • Call direction (inbound or outbound)
  • Agent assigned to the call
No additional configuration is required.

Workflow Triggers

Use the Initiate Call action in your automations to trigger outbound phone calls based on business events:Common Triggers:
  • Record Created (e.g., high-priority support ticket)
  • Record Updated (e.g., order status changed)
  • Scheduled Event (e.g., appointment reminders)
  • Approval Completed (e.g., callback after approval)
  • Time-Based (e.g., daily follow-up calls)
See the Initiating Outbound Calls section for complete configuration details.

Viewing Phone Conversations

Call recordings are not available. To review call content, open the record created for the call and use Chat/Conversation in the record details to view the Voice conversation type, direction (inbound/outbound), and full transcript.

Call Analytics

Use Agent Conversation Ended triggers to build metrics such as call volume (by direction), response times, satisfaction scoring, resolution rates, and agent performance. Capture call data in the trigger, then use AI actions and calculations to produce reports; use the call direction field to segment inbound vs. outbound.

Security and Compliance

  • All phone conversations are processed according to your data retention policies
  • Sensitive information handling follows your organization’s security guidelines
  • Call content is available as transcripts in the Chat/Conversation interface on the call record; call recordings are not accessible to customers

Best Practices

  • Be Concise: Phone conversations require shorter, clearer responses
  • Confirm Information: Always repeat back important details
  • Use Verbal Cues: Include phrases like “Let me help you with that” or “I understand”
  • Provide Clear Next Steps: End calls with specific action items
  • Use endConversation: Train agents to properly close calls using the endConversation tool when resolution is complete
  • Single Language: Use single language mode for better performance unless multilingual support is essential
  • Specific Training: Provide phone-specific training instructions separate from chat training
  • Regular Testing: Periodically test your phone integration to ensure quality
  • Regular Review: Check call logs and agent performance regularly
  • Update Training: Refine agent training based on call outcomes
  • Twilio Management: Monitor Twilio usage and costs
  • Backup Numbers: Consider having backup phone numbers for high-availability scenarios

By following this guide, you can successfully integrate phone functionality with your agents, providing customers with another channel to access your AI-powered support and services.