Skip to main content
Testing is essential before deploying your agent. Atoms provides multiple testing modes so you can verify behavior across different scenarios.
[IMAGE: Test Agent button/modal showing three test mode options]

​
Location

Top right → “Test Agent” button

​
Test Modes

ModeDescriptionBest For
Telephony CallReal phone call to your deviceFull phone experience, final testing
Web CallBrowser voice callQuick voice testing, iterating
ChatText conversationTesting logic only, fast iteration

​
How to Test

  1. Click “Test Agent” button (top right)
  2. Select your testing mode
  3. Start the test
  4. Have a conversation as a real caller would
  5. Try different scenarios and paths

​
What to Test

​
Prompt Understanding

TestWhat to Check
Does the agent understand your questions?Correct interpretation
Are responses appropriate?Relevant to what was asked
Does it follow your prompt instructions?Behavioral alignment

​
Voice Quality

TestWhat to Check
Is speech clear?No audio issues
Natural pacing?Not too fast or slow
Correct pronunciation?Especially for names/terms
Appropriate tone?Matches intended personality

​
End Call Behavior

TestWhat to Check
Say goodbyeDoes call end appropriately?
Confirm resolutionDoes it end when satisfied?
TimeoutDoes it handle silence correctly?

​
Transfer Behavior

TestWhat to Check
Request transferDoes it trigger correctly?
Whisper messageIs context accurate?
Hold experienceDoes music/message play?

​
Variables

TestWhat to Check
Are variables substituted?{{name}} → actual name
Missing variables?Graceful handling

​
Edge Cases

TestWhat to Check
Unexpected inputsGraceful responses
InterruptionsHandles mid-sentence breaks
ConfusionAsks for clarification
Off-topicRedirects appropriately

​
Conversational Flow (if applicable)

TestWhat to Check
Every branchAll paths work
All conditionsCorrect routing
Dead endsNo stuck states
LoopsDon’t trap callers

​
Testing Tips

​
Test Like a Real Caller

Don’t just test happy paths. Real callers will:
  • Interrupt
  • Say unexpected things
  • Ask off-topic questions
  • Be confused
  • Speak unclearly

​
Test All Three Modes

Each mode has different characteristics:
  • Chat is fastest for logic testing
  • Web Call adds voice layer
  • Telephony is most realistic

​
Use Conversation Logs

After each test, review the logs to see:
  • Exact transcripts
  • AI’s interpretation
  • Where things went wrong

​
Iterate

  1. Test
  2. Find issues
  3. Update prompt/config
  4. Test again
  5. Repeat

​
Testing Conversational Flow Agents

For workflow-based agents, test every path: 
Start → Path A → End ✓
Start → Path B → End ✓
Start → Path C → Path D → End ✓
Start → Path C → Path E → End ✓
Start → Fallback → Clarify → Path A ✓
Missing paths are invisible bugs until a real caller hits them.

​
Pre-Deployment Checklist

Before going live:
  • Tested all happy paths
  • Tested error handling
  • Tested edge cases
  • Tested with different voices/accents
  • Verified end call triggers
  • Verified transfer works
  • Reviewed conversation logs
  • Made final prompt adjustments

​
What’s Next

Conversation Logs

Debug with detailed logs

Locking Your Agent

Protect production agents