Skip to main content
Test Call allows you to have voice conversations with your agent before going live. Simulate real calls to verify behavior, voice quality, and conversation flow.

Overview

Test Call provides:
  • Real-time voice interaction with your agent
  • Visual feedback on the canvas (active node highlighting)
  • Conversation transcript
  • Audio quality verification

Starting a Test Call

  1. Open your flow agent in the Flow Builder
  2. Click the Test button in the toolbar
  3. Select Test Call
  4. Allow microphone access when prompted
  5. Start speaking
You’ll need a working microphone and speakers/headphones.

During the Test

Visual Feedback

The canvas shows real-time activity:
IndicatorMeaning
Purple glowCurrently active node
Green edgeTransition being evaluated
Highlighted pathFlow progression

Conversation Panel

The right panel displays:
  • Live transcript of the conversation
  • Agent responses
  • Transition triggers
  • Any extracted data

What to Test

  • Does the voice sound natural?
  • Is the speed appropriate?
  • Are pronunciations correct?
  • Is back-channeling smooth?
  • Do transitions trigger correctly?
  • Does the conversation follow expected paths?
  • Are edge cases handled?
  • Does the agent stay on topic?
  • Are responses accurate?
  • Is the knowledge base being used?
  • Are responses concise?
  • Does the personality match?
  • How long before the agent responds?
  • Is the pacing natural?
  • Does interruption handling work?

Test Scenarios

Run through these common scenarios:

Happy Path

Test the ideal conversation flow:
  1. Start with a greeting
  2. Follow the expected conversation
  3. Complete the intended goal
  4. End the call naturally

Edge Cases

Test unexpected situations:
  • Asking off-topic questions
  • Providing invalid information
  • Requesting human transfer
  • Long pauses/silence

Stress Testing

Push the boundaries:
  • Speak quickly
  • Interrupt the agent
  • Ask complex questions
  • Try to confuse the AI

Reviewing Results

After the test:
  1. Review transcript - Check for accuracy issues
  2. Note timing - Mark slow or awkward moments
  3. Identify failures - Document where flow broke
  4. Test fixes - Make adjustments and retest

Troubleshooting

IssueSolution
No audio from agentCheck speaker settings, verify voice is configured
Agent doesn’t hear meCheck microphone permissions and selection
Delayed responsesNormal for AI processing; check responsiveness setting
Wrong transitionsReview transition conditions
Inaccurate responsesCheck knowledge base attachment and system prompt

Best Practices

Test Incrementally

Test each section as you build, not just at the end.

Document Issues

Keep notes on what doesn’t work so you can fix it.

Test Multiple Times

AI responses vary. Test the same scenario several times.

Use Real Scenarios

Test with questions real customers would ask.