The Four Testing Methods
Dashboard Test Call
Places a real phone call to your configured phone number through the full telephony stack. Tests everything including voice, transcription, LLM, and carrier routing. Use when you want the highest-fidelity test.
Test Console (Simulator)
Simulates call turns in the dashboard without placing a phone call. Faster than a real call for iterating on prompt logic. Does not test voice or transcription.
LLM Playground
Chat-style interface for testing the agent’s LLM behavior in isolation. No voice, no transcription, no telephony. Use for prompt development and function call testing.
Audio Testing
Places a test call specifically to evaluate audio quality. Focus on voice pronunciation, response timing, and denoising behavior.
When to Use Each Method
| Method | Tests | Does Not Test | Best For |
|---|---|---|---|
| Dashboard Test Call | Everything end-to-end | Nothing excluded | Pre-production validation |
| Test Console | LLM behavior, flow logic, turn structure | Voice, transcription, carrier | Rapid prompt iteration |
| LLM Playground | Prompt responses, function calling | Voice, transcription, timing, carrier | Prompt and function development |
| Audio Testing | Voice quality, pronunciation, timing | LLM logic, flow branching | Voice and audio configuration |
Recommended Testing Lifecycle
A structured approach catches different classes of problems at the right stage:Develop the prompt in the LLM Playground
Start with the LLM Playground. Type caller messages and review the agent’s text responses. Iterate on the prompt, system instructions, and function calls until the agent handles your core scenarios correctly.This is the fastest feedback loop. No call setup required.
Validate flow logic in the Test Console
Move to the Test Console to test multi-turn conversations and flow branching. The simulator lets you step through a conversation turn by turn and verify that the agent navigates the flow correctly under different caller inputs.Use this stage to test edge cases: unexpected inputs, early hang-ups, and out-of-scope questions.
Test audio quality
Once prompt behavior is correct, test audio. Place a test call and evaluate voice quality, pronunciation, and response timing. Adjust voice settings in the Audio Tab if needed.Pay particular attention to names, numbers, dates, and domain-specific terminology that text-to-speech engines sometimes mispronounce.
End-to-end test call
Place a full test call through the dashboard. Walk through your core call scenarios as a caller would. Confirm that everything works together: voice recognition, LLM responses, function calls, transfers, and call termination.
Quick Reference
| Method | Where to Find It | Requires Phone? |
|---|---|---|
| Dashboard Test Call | Agent detail > Test Call button | No (calls your configured number) |
| Test Console | Agent detail > Test Console tab | No |
| LLM Playground | Agent detail > LLM Playground tab | No |
| Audio Testing | Agent detail > Test Call > Audio focus | No |
| Simulation Testing | Agent detail > Simulations tab | No |