Skip to main content
After writing your prompt, configure the basic settings that control how your agent sounds, behaves, and interacts with callers.

​
Voice & Language Settings

​
Voice Selection

Choose the voice that will represent your AI during conversations. How to configure:
  1. Navigate to the “Voice Settings” section
  2. Click the voice dropdown
  3. Browse available voices by:
    • Provider: Deepgram, ElevenLabs, PlayHT, etc.
    • Language: Filter by target language
    • Gender: Male, female, or neutral voices
  4. Preview voices before selecting
Best Practices:
  • Match voice to your brand personality
  • Consider your target audience demographics
  • Test voice with actual prompts, not just previews
  • Ensure voice supports your selected language/dialect

​
Language & Dialect

Select the primary language for interactions. Available Options:
  • Language selection with full locale support
  • Dialect-specific options (e.g., US English, UK English, Mexican Spanish)
  • Automatic language detection (optional)
Language Dialect Switcher: Enable this to automatically adapt to the caller’s specific dialect within the same language. Use case: A caller from Scotland speaking English would be better understood with UK English model vs US English.

​
Call Behavior Settings

​
Response Delay

Control how quickly the agent responds after the user stops speaking. Range: 100ms - 1500ms Default: 400ms Lower values (100-300ms):
  • More responsive, faster-paced conversations
  • Better for customer service, sales
  • Risk: May interrupt users who pause naturally
Higher values (600-1500ms):
  • More thoughtful, deliberate responses
  • Better for elderly callers, complex topics
  • Risk: May feel slow or unresponsive
Formula: Each 100ms adjustment changes perceived responsiveness significantly. Recommendation: Start with 400ms, adjust based on testing.

​
Interruption Handling

Choose whether users can interrupt the agent while speaking. Options:
  • Enabled (default): Users can interrupt naturally, like human conversation
  • Disabled: Agent completes full response before listening again
When to disable:
  • Critical information that must be heard completely
  • Legal disclaimers or compliance statements
  • Multi-step instructions
  • Confirmation of important actions
When to enable:
  • Natural customer support conversations
  • Sales calls where engagement is key
  • Interactive Q&A sessions

​
User Inactivity Timeout

Automatically handle callers who don’t respond. Range: 5-30 seconds Default: 15 seconds What happens: After silence, agent will:
  1. Prompt user (“Are you still there?”)
  2. If still no response, either end call or transfer (configurable)
Shorter timeouts (5-10s):
  • Tighter, more efficient conversations
  • Better for high-volume scenarios
Longer timeouts (20-30s):
  • More patient with users who need time to think
  • Better for complex topics or elderly users

​
Max Call Duration

Set maximum call length to prevent runaway costs. Range: 30-600 seconds Default: 300 seconds (5 minutes) What happens: Call automatically ends after maximum duration, with optional farewell message. Recommendations:
  • Customer support: 300-600 seconds
  • Lead qualification: 180-300 seconds
  • Appointment booking: 120-240 seconds
  • Surveys: 120-180 seconds

​
Audio Processing Settings

​
Noise Cancellation

Remove background noise from caller’s audio. Options:
  • Disabled: No noise cancellation (lowest latency)
  • Per Sentence: Processes each sentence separately (balanced)
  • Per Conversation: Processes entire conversation (highest quality)
Latency Impact:
  • Disabled: 0ms
  • Per Sentence: ~50-100ms
  • Per Conversation: ~100-200ms
When to use:
  • Disabled: Quiet environments, latency-critical applications
  • Per Sentence: General use, balanced quality and performance
  • Per Conversation: Noisy environments, quality over speed

​
Background Noise

Add ambient sound to make agent feel more human. Use cases:
  • Call center ambiance for authenticity
  • White noise to avoid “dead air” feeling
  • Environmental sounds for context (office, store, etc.)
Options:
  • Disabled (default): Pure AI voice, no background
  • Enabled: Subtle background ambiance
Note: Don’t combine with noise cancellation—creates processing artifacts.

​
Thinking Voice

Add natural pauses and filler words for human-like conversation. When enabled, agent may use:
  • “Hmm…”
  • “Let me see…”
  • “Ah, okay…”
  • Natural pauses before responses
Benefits:
  • More natural, human-like interaction
  • Gives users time to process information
  • Less jarring during transitions
Drawbacks:
  • Slightly longer conversations
  • May seem less professional in some contexts
Best for: Customer support, friendly conversations Avoid for: Professional/corporate settings, compliance calls

​
Intelligence Features

​
Gender Detection

Automatically detect caller’s gender for personalized responses. How it works: Analyzes voice characteristics to determine likely gender. Use cases:
  • Personalized greetings (“Yes sir” vs “Yes ma’am”)
  • Tailored product recommendations
  • Culturally appropriate responses
Access in prompts: Use the {{user_gender}} variable Privacy note: Ensure compliance with local regulations regarding automated gender detection.

​
Smart Call End

Automatically end calls when conversation naturally concludes. How it works: LLM detects when:
  • User says goodbye or farewell
  • Conversation objectives completed
  • User indicates they’re done
Benefits:
  • Cleaner call endings
  • Better user experience
  • Reduced call duration costs
Note: Combines with max call duration as a safety net.

​
Speaker Identification

Identify different speakers on the same call. Use cases:
  • Speakerphone calls with multiple people
  • Conference calls
  • Household calls where multiple people may speak
Access in prompts: System tracks which speaker said what in conversation history. Limitations: Requires distinct voice characteristics to differentiate speakers.

​
Agentic RAG

Advanced knowledge retrieval with reasoning. Standard RAG: Retrieves relevant documents, passes to LLM Agentic RAG: LLM reasons about what information is needed, retrieves iteratively, synthesizes answer Benefits:
  • More accurate answers to complex questions
  • Better handling of multi-step reasoning
  • Improved context understanding
Drawbacks:
  • Higher latency (additional LLM calls)
  • Increased costs
  • Requires well-structured knowledge base
Best for: Complex technical support, detailed product information, multi-faceted questions

​
Tool Settings

​
Gender Detection (Tool)

Enable gender detection as a callable tool (different from automatic feature). When to use: When you want explicit control over when gender is detected, not automatic.

​
Smart Call End (Tool)

Enable smart call end as a tool the agent can invoke. Difference from setting: Agent decides when to end call based on prompt instructions, rather than automatic detection.

​
Example Configurations

​
Friendly Customer Support

Voice: Natural, conversational female voice
Response Delay: 400ms
Interruption: Enabled
Inactivity Timeout: 20s
Max Duration: 600s
Noise Cancellation: Per Sentence
Thinking Voice: Enabled
Smart Call End: Enabled

​
Professional Corporate Helpdesk

Voice: Clear, professional neutral voice
Response Delay: 500ms
Interruption: Disabled for critical info
Inactivity Timeout: 15s
Max Duration: 480s
Noise Cancellation: Per Conversation
Thinking Voice: Disabled
Smart Call End: Enabled

​
Fast-Paced Sales

Voice: Energetic, confident voice
Response Delay: 200ms
Interruption: Enabled
Inactivity Timeout: 10s
Max Duration: 300s
Noise Cancellation: Per Sentence
Thinking Voice: Disabled
Smart Call End: Disabled (agent controls end)

​
Elderly Care Support

Voice: Warm, slow-paced voice
Response Delay: 800ms
Interruption: Disabled
Inactivity Timeout: 30s
Max Duration: 600s
Noise Cancellation: Per Conversation
Thinking Voice: Enabled
Smart Call End: Enabled

​
Testing Your Configuration

After configuring settings:
  1. Test in browser first with various scenarios
  2. Test via phone call to verify audio quality
  3. Test interruptions by speaking over agent
  4. Test silence handling by staying quiet
  5. Test noise in different environments
  6. Test with real users from your target demographic

​
Next Steps

Ready for complex flows? Explore Flow Agent Global Settings for node-based configuration.