Skip to main content
Thorough testing ensures your AI agents perform correctly before going live. Emanate provides built-in test panels for both voice and chat agents.

​
Test Panel Overview

Every agent has a test panel accessible from the editor:
  1. Open the agent in the editor
  2. Click the Test button
  3. Start a test session

​
Voice Test Panel

The voice test panel simulates phone calls:
  • Start Call: Begin a test voice session
  • End Call: Terminate the session
  • Transcript: Real-time conversation display
  • Debug View: Tool invocations and model responses

​
Chat Test Panel

The chat test panel simulates widget conversations:
  • Message Input: Type test messages
  • Conversation View: See agent responses
  • Debug Panel: View underlying processing

​
Testing Workflow

​
1. Basic Functionality

Start with core functionality:
TestVerify
First messageGreeting plays/displays correctly
Simple questionAgent responds appropriately
Follow-upContext is maintained
GoodbyeSession ends gracefully

​
2. Knowledge Base

Test document retrieval: 
You: "What products do you offer?"
Agent: [Should reference knowledge base documents]

You: "Tell me about SKU-1234"
Agent: [Should find specific product info]

​
3. Lead Capture

Verify data collection: 
You: "I'm interested in getting a quote"
Agent: "I'd be happy to help. May I have your name?"
You: "John Smith"
Agent: "Thanks John. What's the best email to reach you?"
You: "john@company.com"
[Verify lead record is created]

​
4. Custom Actions

Test tool invocations: 
You: "Do you have 100 units of steel beams in stock?"
Agent: [Should invoke inventory check tool]
Agent: "Yes, we have 250 units available..."

​
5. Edge Cases

Test unusual scenarios:
ScenarioExpected Behavior
Silence (5+ seconds)Agent prompts user
InterruptionAgent handles gracefully
Off-topic questionRedirects to scope
Angry customerDe-escalation response
Transfer requestInitiates handoff

​
Voice-Specific Testing

​
Audio Quality

TestWhat to Check
Speech recognitionAgent understands clearly
Voice outputNatural, correct pacing
Background noiseHandles ambient sound

​
Conversation Flow

  • Barge-in: Can you interrupt the agent?
  • Turn-taking: Natural pauses between speakers?
  • Long responses: Agent breaks up appropriately?

​
Phone Features

FeatureTest
Hold”Can you hold for a moment?”
Transfer”Let me speak to a human”
VoicemailCall when unavailable
RecordingVerify calls are recorded

​
Chat-Specific Testing

​
Widget Functionality

TestVerify
Widget loadsAppears on page correctly
Open/closeToggle works
MinimizeStays accessible
MobileResponsive on small screens

​
Message Handling

  • Long messages: Agent handles appropriately
  • Quick succession: Multiple messages in a row
  • Links: Clickable and correct
  • Formatting: Markdown renders properly

​
Session Persistence

  1. Start a conversation
  2. Navigate to another page
  3. Return to original page
  4. Verify conversation continues

​
Reviewing Test Results

The debug view shows the model’s input and output for each turn, so you can see exactly what the agent received and how it responded. You can see which tools were called, what information was sent, and what response came back. Lead capture data is displayed so you can verify the agent collected the right information.

​
Test Scenarios

​
Sales Qualification

## Scenario: Qualified Lead
1. Caller asks about products
2. Agent provides information
3. Caller expresses interest in quote
4. Agent collects: name, email, company, requirements
5. Agent offers to schedule meeting
6. Lead captured with all fields

## Expected: Lead created with ICP score > 70

​
Technical Support

## Scenario: Known Issue
1. Caller reports problem
2. Agent identifies issue from knowledge base
3. Agent provides solution
4. Caller confirms resolution
5. Session ends positively

## Expected: No escalation, issue resolved

​
Escalation

## Scenario: Complex Request
1. Caller has unusual requirement
2. Agent cannot find answer in knowledge base
3. Agent offers to transfer to specialist
4. Transfer executes correctly

## Expected: Smooth handoff, context preserved

​
Automated Testing

​
Test Scripts

Create reusable test scenarios with expected responses. Define a series of messages and what you expect the agent to say or do.

​
Regression Testing

Run tests after changes:
  1. Save test scenarios
  2. Make agent changes
  3. Run all scenarios
  4. Compare results to baseline

​
Pre-Deployment Checklist

Before going live:

​
Agent Configuration

  • System prompt is complete
  • First message is appropriate
  • Model settings are optimal
  • Voice sounds natural (if voice)

​
Knowledge Base

  • All documents uploaded
  • Documents are processed
  • Test queries return correct info

​
Custom Actions

  • All tools configured
  • API endpoints working
  • Error handling in place

​
Lead Capture

  • Fields configured correctly
  • Data saves to database
  • Enrichment triggers (if enabled)

​
Edge Cases

  • Handles silence/pauses
  • Handles interruptions
  • Handles off-topic questions
  • Transfer/escalation works

​
Compliance

  • Recording disclosure (if required)
  • Data handling compliant
  • Boundaries respected

​
Troubleshooting Test Issues

  • Check browser microphone permissions
  • Verify internet connection
  • Try a different browser
  • Check system prompt isn’t empty
  • Verify model is configured
  • Look for errors in debug view
  • Verify tool configuration
  • Check API endpoints are accessible
  • Review tool invocation in debug view

​
Next Steps

Voice Agents

Voice configuration

Chat Agents

Chat configuration

Evaluations

Quality assessment

Deploy Campaign

Launch outbound calls