Skip to main content

Configure your first AI phone agent

Complete guide to create, configure, test, and publish an AI phone agent that handles customer conversations automatically. In this guide, you'll set up a professional voice agent that can make sales calls, provide customer support, schedule appointments, or handle any other voice interaction use case.

Overview​

An AI agent in Agoralia is a voice-enabled AI assistant that can:

  • Answer inbound calls automatically
  • Make outbound calls for campaigns
  • Handle conversations in natural language
  • Extract structured data from conversations
  • Integrate with knowledge bases for enhanced responses
  • Use tools and functions for external integrations

Agents are configured through a 5-step wizard that guides you through all settings, from basic configuration to advanced features.

Prerequisites​

Before creating an agent, ensure you have:

  • ✅ Admin or editor access to your Agoralia workspace
  • ✅ Clear understanding of your use case (sales, support, appointments, etc.)
  • ✅ Optional: Knowledge base documents prepared (for enhanced responses)
  • ✅ Optional: Phone number configured (for testing)

Step 1: Base Configuration​

The first step sets up the fundamental properties of your agent.

Agent Name​

Required: Give your agent a descriptive name that reflects its purpose.

Examples:

  • "Enterprise Sales Rep"
  • "Tech Support Assistant"
  • "Appointment Scheduler"
  • "Customer Onboarding Agent"

Best Practices:

  • Use clear, descriptive names
  • Include the use case or department
  • Avoid generic names like "Agent 1"

Language​

Required: Select the primary language for your agent.

Available Languages: Agoralia supports 150+ languages including:

  • English (US, UK, Canada, Australia, etc.)
  • Italian, Spanish, French, German
  • Chinese, Japanese, Korean
  • Arabic, Hindi, Portuguese
  • And many more...

How it works:

  • Agoralia automatically selects the best voice engine for your chosen language
  • English languages use optimized voice engines
  • All other languages are fully supported with high-quality voices

Note: You can enable multilingual mode in campaigns to auto-detect the caller's language.

Voice ID​

Required: Select the voice that will speak for your agent.

Voice Selection:

  • High-quality voices are available for all supported languages
  • Voices are automatically filtered based on your selected language
  • You can preview voices before selecting them

Voice Properties:

  • Gender: Male, Female, Neutral
  • Accent: Regional variations (e.g., US English vs. UK English)
  • Style: Professional, Casual, Friendly, etc.

Preview: You can preview voices before selecting them using the voice selector dialog.

Best Practices:

  • Choose a voice that matches your brand personality
  • Test multiple voices to find the best fit
  • Consider your target audience's preferences

LLM Model​

Required: Select the Large Language Model (LLM) that powers your agent's intelligence.

Available Models:

  • GPT-4o: Most capable, best for complex conversations
  • GPT-4o-mini: Faster and more cost-effective, good for most use cases
  • GPT-3.5-turbo: Legacy model, still supported
  • Additional models may be available based on your plan

Model Selection Guide:

  • Complex use cases (technical support, sales): Use GPT-4o
  • Simple use cases (appointments, FAQs): Use GPT-4o-mini
  • Cost-sensitive: Use GPT-4o-mini
  • Speed-critical: Use GPT-4o-mini

Note: Available models may vary. Contact support if you need a specific model.

Step 2: Character Configuration​

Define your agent's personality, role, and conversation style.

Primary Role​

Options:

  • Inbound: Agent handles incoming calls only
  • Outbound: Agent makes outbound calls only
  • Both: Agent handles both inbound and outbound calls

Default: both

Use Cases:

  • Inbound only: Customer support, help desk
  • Outbound only: Sales outreach, appointment reminders
  • Both: General-purpose agents

Mission​

Required: Describe what your agent should accomplish.

Examples:

  • "Help customers troubleshoot technical issues with our product"
  • "Schedule appointments for our dental practice"
  • "Qualify leads for our sales team"
  • "Provide information about our services and pricing"

Best Practices:

  • Be specific and clear
  • Include the primary goal
  • Mention any constraints or requirements
  • Keep it concise (1-2 sentences)

Additional Instructions (Custom Prompt)​

Optional: Provide additional instructions for the agent's behavior.

Use Cases:

  • Tone and style guidelines
  • Specific phrases to use or avoid
  • Handling of edge cases
  • Compliance requirements

Example:

Always be polite and professional. If the customer seems frustrated, 
acknowledge their feelings and offer to escalate to a human agent. 
Never make promises about pricing or delivery dates without checking 
our system first.

Welcome Message​

Optional: Custom greeting message that the agent says when answering a call.

Examples:

  • "Hello! Thank you for calling Acme Corp. How can I help you today?"
  • "Hi, this is Sarah from Tech Support. What seems to be the issue?"
  • "Welcome to our appointment booking service. I'm here to help you schedule your visit."

If not provided: The agent will generate a greeting based on its mission.

Who Speaks First​

Options:

  • Agent: Agent greets the caller first (default)
  • User: Agent waits for the caller to speak first

Default: agent

Use Cases:

  • Agent first: Most common, provides immediate greeting
  • User first: Useful for inbound calls where caller might have urgent questions

Pause Before Speaking​

Range: 0-5000 milliseconds

Default: 0ms

Purpose: Adds a delay before the agent starts speaking (useful for ensuring the call is connected).

Recommendations:

  • 0ms: For most use cases
  • 500-1000ms: If you experience connection issues
  • 2000-5000ms: For very slow connections (not recommended)

Step 3: Voice & Behavior​

Fine-tune how your agent sounds and behaves during conversations.

Voice Settings​

Voice Model​

Optional: Advanced voice model configuration (usually auto-detected from voice ID).

Voice Temperature​

Range: 0.0 - 2.0

Default: 1.0

Effect: Controls the randomness and creativity of the voice.

  • Lower (0.0-0.7): More consistent, predictable voice
  • Higher (1.3-2.0): More varied, expressive voice

Recommendations:

  • Professional/Formal: 0.7-1.0
  • Friendly/Casual: 1.0-1.3
  • Creative/Expressive: 1.3-1.7

Voice Speed​

Range: 0.5 - 2.0

Default: 1.0

Effect: Controls how fast the agent speaks.

  • Slower (0.5-0.8): Easier to understand, good for complex information
  • Normal (0.9-1.1): Standard speaking pace
  • Faster (1.2-2.0): Quick delivery, use sparingly

Recommendations:

  • Technical support: 0.8-1.0 (slower for clarity)
  • Sales: 1.0-1.2 (energetic but clear)
  • Appointments: 0.9-1.1 (balanced)

Volume​

Range: 0.0 - 2.0

Default: 1.0

Effect: Controls the audio volume level.

Recommendations:

  • Most cases: 1.0 (default)
  • Quiet environments: 1.2-1.5
  • Noisy environments: 0.8-1.0

Behavior Settings​

Responsiveness​

Range: 0.0 - 1.0

Default: 1.0

Effect: How quickly the agent responds to the user's speech.

  • Lower (0.0-0.5): Waits longer before responding (more natural pauses)
  • Higher (0.7-1.0): Responds quickly (more conversational)

Recommendations:

  • Formal/Professional: 0.6-0.8
  • Casual/Friendly: 0.8-1.0

Interruption Sensitivity​

Range: 0.0 - 1.0

Default: 1.0

Effect: How easily the agent allows the user to interrupt.

  • Lower (0.0-0.5): Agent continues speaking, harder to interrupt
  • Higher (0.7-1.0): Agent stops quickly when user speaks

Recommendations:

  • Information-heavy: 0.5-0.7 (less interruptions)
  • Conversational: 0.8-1.0 (more interruptions allowed)

Enable Backchannel​

Default: Disabled

Effect: Agent makes small verbal acknowledgments ("uh-huh", "I see", "okay") during user speech.

Use Cases:

  • Enabled: More natural, conversational feel
  • Disabled: More formal, professional feel

When Enabled:

Backchannel Frequency​

Range: 0.0 - 1.0

Default: 0.8

Effect: How often the agent makes backchannel sounds.

Normalize for Speech​

Default: Disabled

Effect: Automatically converts text to speech-friendly format (e.g., "Dr." → "Doctor", "$100" → "one hundred dollars").

Use Cases:

  • Enabled: When agent reads structured data (dates, numbers, abbreviations)
  • Disabled: For natural conversation

Background Sound (Ambient Sound)​

Options:

  • None (default)
  • Coffee Shop: Subtle café ambiance
  • Convention Hall: Large room reverb
  • Summer Outdoor: Outdoor environment
  • Mountain Outdoor: Natural outdoor setting
  • Static Noise: Background static
  • Call Center: Office/call center ambiance

Use Cases:

  • None: Most professional use cases
  • Coffee Shop: More casual, friendly feel
  • Call Center: Traditional call center experience

When Enabled:

Ambient Sound Volume​

Range: 0.0 - 2.0

Default: 1.0

Effect: Volume of the background sound relative to the agent's voice.

Recommendations: Keep below 0.5 to avoid distracting from the conversation.

Step 4: Transcription & Call Settings​

Configure how the agent processes speech and manages calls.

Transcription Settings​

Transcription Mode​

Options:

  • Fast: Lower latency, good for real-time conversations (default)
  • Accurate: Higher accuracy, slightly higher latency

Recommendations:

  • Fast: For most use cases, real-time conversations
  • Accurate: For technical terms, medical terminology, or when accuracy is critical

Vocabulary Specialization​

Options:

  • General (default): Standard vocabulary
  • Medical: Optimized for medical terminology

Use Cases:

  • General: Most use cases
  • Medical: Healthcare, medical appointments, pharmaceutical sales

Denoising Mode​

Options:

  • Noise Cancellation: Removes background noise only
  • Noise and Background Speech Cancellation: Removes both noise and background speech

Recommendations:

  • Noise Cancellation: For most environments
  • Noise and Background Speech: For noisy call centers or open offices

Boosted Keywords​

Optional: Comma-separated list of keywords that should be recognized with higher priority.

Format: keyword1, keyword2, keyword3

Examples:

  • Product names: AcmePro, Enterprise Edition, Premium Plan
  • Technical terms: API, SDK, webhook, integration
  • Industry terms: GDPR, HIPAA, compliance

Use Cases:

  • Product names or brand names
  • Technical terminology
  • Industry-specific terms
  • Acronyms

Call Settings​

End Call After Silence​

Range: 10 seconds minimum

Default: 600 seconds (10 minutes)

Effect: Automatically ends the call if there's no speech for this duration.

Recommendations:

  • Short calls (appointments, FAQs): 30-60 seconds
  • Medium calls (support, sales): 60-120 seconds
  • Long calls (consultations): 300-600 seconds

Maximum Call Duration​

Range: 60 seconds (1 minute) - 7200 seconds (2 hours)

Default: 3600 seconds (60 minutes)

Effect: Maximum length of a call before automatic termination.

Recommendations:

  • Appointments: 300-600 seconds (5-10 minutes)
  • Support: 600-1800 seconds (10-30 minutes)
  • Sales: 1800-3600 seconds (30-60 minutes)
  • Consultations: 3600-7200 seconds (60-120 minutes)

Ring Duration​

Range: 5-90 seconds

Default: 30 seconds

Effect: How long the phone rings before the call is considered unanswered.

Recommendations:

  • Most cases: 30 seconds
  • High-volume: 20 seconds
  • Important calls: 60-90 seconds

DTMF (Touch-Tone) Support​

Default: Enabled

Effect: Allows users to press numbers on their phone keypad to interact with the agent.

Use Cases:

  • Menu navigation ("Press 1 for Sales, Press 2 for Support")
  • Confirmation ("Press 1 to confirm")
  • Data entry (phone numbers, account numbers)

When Enabled: You can configure DTMF options in the agent's advanced settings.

Voicemail Detection​

Default: Disabled

Effect: Automatically detects when a call goes to voicemail and leaves a message.

When Enabled:

Voicemail Message​

Optional: Custom message to leave on voicemail.

Example:

Hi, this is Sarah from Acme Corp. I'm calling to follow up on your 
inquiry. Please call us back at 1-800-555-1234 or visit our website 
at acme.com. Thank you!

If not provided: The agent will generate a message based on its mission.

Step 5: Advanced & Knowledge Base​

Configure advanced features like knowledge bases, webhooks, tools, and integrations.

Knowledge Base​

Optional: Connect your agent to knowledge bases for enhanced responses.

How it works:

  • Agent can reference knowledge base content during conversations
  • Knowledge bases can contain FAQs, product information, policies, etc.
  • Multiple knowledge bases can be connected to a single agent

Available Knowledge Bases:

  • General KB: Automatically available to all agents
  • Custom KBs: Created in the Knowledge Base section

Selection:

  • Check the boxes next to knowledge bases you want to connect
  • Knowledge bases are filtered by language (agent language must match KB language)
  • Synced knowledge bases are marked with "✓ Synced"

Best Practices:

  • Connect relevant knowledge bases for your use case
  • Keep knowledge bases up-to-date
  • Use language-specific knowledge bases for multilingual agents

Webhook Settings​

Optional: Configure webhooks to receive call events in real-time.

Webhook URL​

Format: Valid HTTPS URL

Example: https://your-backend.com/webhooks/agoralia

Events Sent:

  • Call started
  • Call ended
  • Call transcript updates
  • Extracted data
  • Call outcomes

Use Cases:

  • Real-time call monitoring
  • CRM integration
  • Custom analytics
  • Alert systems

Webhook Timeout​

Range: 0+ milliseconds

Default: 5000ms (5 seconds)

Effect: Maximum time to wait for webhook response before timing out.

Recommendations:

  • Fast webhooks: 3000-5000ms
  • Slow webhooks: 10000-30000ms

Tools & Functions​

Optional: Add custom tools/functions that the agent can call during conversations.

Available Tools:

  • Custom Functions: Define your own functions
  • MCP Servers: Connect to Model Context Protocol servers
  • API Integrations: Connect to external APIs

Use Cases:

  • Calendar Integration: Check availability, book appointments
  • CRM Integration: Look up customer data, update records
  • Database Queries: Retrieve information
  • API Calls: External service integration

Configuration:

  • Tools are configured in the Functions Editor
  • Each tool has a name, description, and parameters
  • Agent automatically decides when to use tools based on conversation context

MCP Servers​

Optional: Connect Model Context Protocol (MCP) servers for advanced integrations.

MCP Servers:

  • Provide structured access to external data and functions
  • Enable agents to interact with complex systems
  • Support authentication and secure connections

Configuration:

  • Select MCP servers from your configured servers
  • Servers must be set up in Settings → Integrations → MCP Servers
  • Agent can use multiple MCP servers simultaneously

Post-Call Analysis​

Optional: Configure automatic analysis of call transcripts after calls end.

Analysis Types:

  • Sentiment Analysis: Determine caller sentiment
  • Topic Extraction: Identify main topics discussed
  • Action Items: Extract follow-up actions
  • Summary Generation: Create call summaries

Model: Select the LLM model to use for analysis (default: GPT-4o-mini)

Use Cases:

  • Quality assurance
  • Call categorization
  • Performance metrics
  • Training data generation

Data Storage​

Options:

  • Everything: Store all call data including transcripts and audio
  • Everything Except PII: Store data but redact personally identifiable information
  • Basic Attributes Only: Store only metadata (duration, outcome, etc.)

Default: Everything

Recommendations:

  • GDPR/Privacy Compliance: Use "Everything Except PII" or "Basic Attributes Only"
  • Full Analytics: Use "Everything"
  • Cost Optimization: Use "Basic Attributes Only"

Opt-In Signed URL​

Default: Disabled

Effect: Generate signed URLs for accessing call recordings (requires opt-in).

Use Cases:

  • Secure access to call recordings
  • Time-limited access links
  • Compliance with data retention policies

Testing Your Agent​

Before publishing, test your agent to ensure it works correctly.

Test Call​

How to Test:

  1. Go to the agent's edit page
  2. Click "Test Call" button
  3. Enter a phone number (your own number for testing)
  4. Click "Start Test Call"
  5. Answer the call and have a conversation

What to Test:

  • ✅ Agent answers correctly
  • ✅ Voice quality is good
  • ✅ Agent understands your questions
  • ✅ Agent responds appropriately
  • ✅ Agent follows instructions correctly
  • ✅ Call ends properly

Test Chat (Simulation)​

How to Test:

  1. Go to the agent's edit page
  2. Click the "Simulation" tab
  3. Type messages as if you were calling
  4. Review the agent's responses

What to Test:

  • ✅ Agent understands context
  • ✅ Responses are relevant
  • ✅ Agent follows instructions
  • ✅ Knowledge base content is used correctly
  • ✅ Tools/functions are called when needed

Cost Estimate​

How to View:

  • Cost estimate is shown automatically when configuring the agent
  • Based on selected language, voice, and LLM model
  • Updates in real-time as you change settings

What's Included:

  • Per-minute rates for voice
  • Per-minute rates for LLM
  • Estimated cost per call
  • Estimated cost per 1000 calls

Publishing Your Agent​

Once your agent is configured and tested, publish it to make it available for campaigns.

How to Publish​

  1. Complete all 5 steps of the agent configuration
  2. Review all settings
  3. Click "Save" or "Publish" button
  4. Agent is now available for use in campaigns

Agent Status​

Draft: Agent is saved but not published (can be edited) Published: Agent is available for campaigns (can still be edited)

Note: Changes to published agents take effect immediately for new calls. Existing calls continue with the previous configuration.

Best Practices​

Agent Configuration​

  • ✅ Start simple: Begin with basic settings, add complexity as needed
  • ✅ Test thoroughly: Test with real calls before deploying to production
  • ✅ Monitor performance: Review call transcripts and adjust settings
  • ✅ Iterate: Improve agents based on real-world usage

Voice Selection​

  • ✅ Match brand: Choose voices that match your brand personality
  • ✅ Consider audience: Test with your target audience
  • ✅ Test multiple: Try different voices to find the best fit

Language & Model​

  • ✅ Use appropriate model: GPT-4o for complex, GPT-4o-mini for simple
  • ✅ Consider costs: Balance capability with cost
  • ✅ Test performance: Ensure model handles your use case well

Knowledge Bases​

  • ✅ Keep updated: Regularly update knowledge base content
  • ✅ Be specific: Include detailed, accurate information
  • ✅ Test relevance: Ensure KB content is actually used

Webhooks & Integrations​

  • ✅ Handle errors: Implement proper error handling in webhooks
  • ✅ Test endpoints: Verify webhooks work before deploying
  • ✅ Monitor performance: Track webhook response times

Troubleshooting​

Agent Doesn't Answer Calls​

Possible Causes:

  • Agent is not published
  • Phone number not configured correctly
  • Agent configuration incomplete

Solutions:

  • Verify agent status is "Published"
  • Check phone number configuration
  • Test agent using "Test Call" feature

Poor Voice Quality​

Possible Causes:

  • Voice selection mismatch
  • Network issues
  • Voice settings configuration

Solutions:

  • Try different voice
  • Check network connection
  • Adjust voice settings (speed, volume)

Agent Doesn't Understand Questions​

Possible Causes:

  • Language mismatch
  • Transcription issues
  • LLM model limitations

Solutions:

  • Verify language is correct
  • Check transcription settings
  • Try a more capable LLM model (GPT-4o)

Knowledge Base Not Used​

Possible Causes:

  • Knowledge base not connected
  • Language mismatch
  • Content not relevant

Solutions:

  • Verify knowledge base is selected in Step 5
  • Ensure KB language matches agent language
  • Review KB content for relevance

Next Steps​

Once your agent is configured and published:

Additional Resources​

Was this page helpful? Yes | No