Observability
Monitor, debug, and optimize your deployed models in real-time.
Real-time metrics
View latency (p50, p95, p99), throughput, error rates, and GPU utilization in the dashboard. Metrics update every second with 30-day retention.
Request tracing
Trace individual requests end-to-end with unique request IDs. See preprocessing time, inference time, and network latency breakdowns.
Log streaming
Stream logs in real-time with `upbox logs --follow` or forward to Datadog, Splunk, or CloudWatch with one-click integrations.
Alerting
Configure alerts for latency spikes, error rate thresholds, or traffic anomalies. Notifications via Slack, PagerDuty, email, or webhooks.
Model drift detection
Upbox monitors input distributions and flags drift automatically. Set up alerts when predictions shift or input patterns change unexpectedly.
OpenTelemetry export
Export traces and metrics to any OpenTelemetry-compatible backend. Enable with `upbox config set telemetry.otlp_endpoint <url>`.