Skip to main content
Complete these checks before enabling your agent for production traffic.
Never enable an agent for production without completing all Critical items.

​
Critical checks

​
Agent configuration

  • System prompt is production-ready (no placeholders or TODOs)
  • Agent name and description are accurate (not “Test Agent”)
  • Agent is currently disabled (isEnabled: false)

​
LLM and voice

  • Model latency tested under 2 seconds
  • Temperature set between 0.6-0.8
  • TTS voice pronunciation verified for key terms
  • Speech recognition tested with target accents

​
Tools and integrations

  • All tools tested in actual conversations
  • Tool endpoints are production URLs (not localhost/staging)
  • Tool authentication uses production credentials
  • All URLs use HTTPS

​
Security

  • Production API keys configured and stored securely
  • Web integration origins restricted (no wildcard *)
  • Webhook endpoints use HTTPS with authentication

​
Testing completed

  • Core use case tested 10+ times with voice
  • Multi-turn conversations maintain context
  • Silence handling works (agent prompts after 5+ seconds)
  • Interruption handling works (agent stops and listens)

​
Monitoring

  • Dashboard access verified for team
  • Error alerts configured (alert when error rate > 5%)
  • On-call rotation defined with escalation path

​
Important checks

  • System prompt reviewed by team member
  • Edge cases tested (unclear input, out-of-scope requests)
  • Tool timeout handling verified
  • Support team trained on agent capabilities
  • Incident response runbook created

​
Launch procedure

​
Pre-launch (1 hour before)

  1. Verify all critical items completed
  2. Run final end-to-end test
  3. Confirm team is available and briefed
  4. Confirm rollback plan is ready

​
Go-live

  1. Enable agent (isEnabled: true)
  2. Review first 5 calls in real-time
  3. Verify no errors or unexpected behavior

​
First hour

  • Active monitoring by team member
  • Check error rate, latency, and webhook status
  • Review early transcripts for issues

​
Key metrics

​
When to disable immediately

  • Error rate > 20% for 10+ minutes
  • Security breach suspected
  • Agent providing harmful responses
  • Unexpected cost spike

​
Emergency rollback

  1. Disable agent — Set isEnabled to false
  2. Notify stakeholders — Inform team of incident
  3. Preserve evidence — Export transcripts and logs
  4. Investigate — Review recent changes
  5. Fix and re-test — Before re-enabling
Know how to disable your agent before going live. Practice the rollback procedure with your team.

Call History

Monitor agent performance

Call Inspector

Debug specific calls