AI Monitoring
Real-time visibility into agent performance, runtime health and system behaviour.
AI monitoring in Command Center provides real-time visibility into every running agent and pipeline. It tracks latency, error rates, cost, quality and health status — surfacing problems before they reach users. Without monitoring, production AI is a black box: you deploy and hope. With monitoring, you operate with confidence.
Key Metrics
- Latency — p50 and p95 response time per agent. Spikes indicate model degradation or capacity issues.
- Error rate — percentage of failed executions. Increases indicate integration failures or model errors.
- Cost — per-call and aggregate cost per agent, pipeline and workspace. Prevents budget overruns.
- Quality — accuracy and confidence scores over time. Degradation indicates drift or input distribution shift.
- Throughput — requests per minute. Capacity planning and bottleneck identification.
Health Status
Every agent has a health status: healthy, degraded or failing. The status is derived from real-time metrics — if error rate exceeds 5% or p95 latency exceeds 2 seconds, the status changes to degraded. If error rate exceeds 20%, the status is failing. Health status drives alerting and auto-rollback.
Monitoring is not just for operators. Auditors need historical monitoring data to verify that the system was operating correctly during a specific period. Command Center retains monitoring data for 13 months at aggregated resolution, satisfying most compliance requirements.
FAQ
Frequently Asked Questions
Metrics are aggregated and displayed within seconds of the event. Dashboards refresh automatically. Alerts fire within 30 seconds of a threshold breach. There is no batch processing delay — the data you see is the current state.
Yes. Command Center supports custom dashboards with configurable widgets. You can create role-specific views: an operations dashboard for on-call engineers, an executive dashboard for leadership and a compliance dashboard for auditors.
Related Topics