How to Test Agents
Never launch untested.
Use the Playground, Simulation Mode, and Live Preview to ensure your AI sounds natural, handles objections, and books appointments — all without spending credits.
1. Open the Testing Playground
- Go to Agents → Select your agent → Click Test in Playground
- Or: Dashboard → Playground → Drag your agent in
2. Test Modes
| Mode | Purpose | Cost |
|---|
| Text Simulation | Type user replies | Free |
| Voice Input | Speak into mic | Free |
| Audio Upload | Test with real call recordings | Free |
| Live Preview | Call your agent from your phone | 1 credit |
3. Run a Full Test Script
Example: Dental Booking Agent
| Step | You Say/Type | Expected AI Response |
|---|
| 1 | ”Hi, I need to book an appointment" | "Great! May I have your full name?“ |
| 2 | ”John Smith" | "Thanks, John. Are you a new or returning patient?“ |
| 3 | ”New" | "Welcome! When are you available next week?“ |
| 4 | ”Tuesday at 3pm" | "Perfect. I’ve booked you with Dr. Lee on Tuesday at 3pm. You’ll get a confirmation text.” |
After each test, get an AI Scorecard:
| Metric | Score | Notes |
|---|
| Naturalness | 9.2/10 | Pauses perfect |
| Goal Completion | 100% | Booking made |
| Objection Handling | 8.5/10 | Could improve “too busy” |
| Latency | 420ms | Fast |
5. Debug & Fix
- Red highlights = AI failed or went off-script
- Click any line → Edit Prompt → Fix logic
- Re-run test instantly
6. Client Handoff (Agency Workflow)
- Clone tested agent →
ClientX - Final
- Run client-specific test script (their name, number, brand)
- Record demo call in Playground
- Click Share Demo → Send secure link to client
- Deploy to Client Portal → They see “Your AI is live and tested”
Always test 3 scenarios:
- Ideal path
- Objection (“I’m busy”)
- Edge case (wrong input)
All tests are private — no data leaves your account.
