Testing workflow
Phase 1: Configuration
1
Create disabled agent
Set
isEnabled: false and configure LLM, voice, and system prompt2
Preview voice
Compare voices, adjust speed (0.9-1.1x recommended)
3
Test prompts
Use chat mode to iterate on prompts quickly
4
Test webhooks
Verify endpoints return HTTP 200
Phase 2: Conversation testing
1
Test happy paths
Run ideal conversation flows
2
Test edge cases
Try unclear input, interruptions, out-of-scope requests
3
Test errors
Disconnect webhooks, test timeout handling
4
Voice quality
Switch to voice mode, check clarity and latency
Phase 3: Integration testing
1
Test tools
Trigger each tool, verify parameter extraction
2
Test MCP connections
Validate server connectivity
3
Test schedule
Verify business hours behavior
4
Test concurrency
Schedule multiple calls via API
Phase 4: Pre-production
1
End-to-end voice test
Test all conversation branches
2
Review metrics
Target: <2 seconds latency
3
Team review
Gather feedback on conversation quality
4
Enable for production
Monitor first 10-20 calls in Call Inspector
Checklists
Essential (every agent)
- Voice sounds natural and clear
- Core flows complete successfully
- System prompt instructions followed
- 10+ scenarios validated
- Error handling tested
Advanced (complex agents)
- All tools execute successfully
- Webhooks deliver and retry
- MCP servers functional
- Multi-turn coherence maintained
Edge cases (production-critical)
- Long silences handled (5+ seconds)
- Background noise tolerated
- Out-of-scope requests declined gracefully
What’s next
Dashboard Testing
Test agents in the browser
Call Inspector
Analyze completed calls in detail