📄 中文摘要
大多数 VAPI 部署在出现问题时往往默默失败，例如通话中断、延迟激增和 webhook 超时，直到客户投诉才被发现。有效的监控应关注三个方面：通话成功率（HTTP 200 与 5xx 的对比）、语音识别和语音合成的延迟（目标小于 500 毫秒）、以及 webhook 的交付情况（重试次数和超时模式）。将 Twilio 的错误回调集成到可观察性堆栈中，追踪语音活动检测的误报和插话竞争条件。实时警报可以帮助及时发现通话放弃率的基础设施问题，防止问题扩散。
监控 VAPI 部署：有效的性能跟踪与错误报告
📄 English Summary
Monitoring VAPI Deployments: Effective Performance Tracking and Error Reporting
Many VAPI deployments fail silently, with issues like dropped calls, latency spikes, and webhook timeouts going unnoticed until customers raise complaints. Effective monitoring should focus on three key areas: call success rates (comparing HTTP 200 to 5xx), STT/TTS latency (targeting under 500ms), and webhook delivery metrics (including retry counts and timeout patterns). Integrating Twilio's error callbacks into the observability stack is essential for tracking VAD false positives and barge-in race conditions. Real-time alerts on call abandonment rates can help identify infrastructure issues before they escalate.