Skip to content

Performance

Performance reports org/service-level throughput, throttling, and quality-lens latency for your traffic — distinct from pod-level infrastructure metrics, which are an operator concern and aren’t shown here. It includes a summary status bar, throughput/throttling and latency trend charts, a per-model latency table, and — where the platform read is reachable — real per-model p50/p95/p99 request-latency percentiles.

The Performance panel showing a summary status bar, throughput and throttling KPI tiles, a throughput line chart, a model-latency line chart, and a per-model latency percentiles table.
Monitoring & Reporting → PerformanceCitadel synthetic data on a non-production Arbitex demo instance. Never client data.
  1. Open Monitoring & Reporting → Performance. Pick a Range for the charts and an Auto-refresh cadence, or click Refresh.

  2. Review the summary status bar (Throughput, Throttling, Latency anomalies, Overall) and the KPI tiles (Latency anomalies, Peak requests, Throttled, Avg throttle rate).

  3. Review Throughput & throttling and Model latency / quality lenses trend charts, with anomaly badges on flagged metrics.

  4. Review Latency by model — the latest value, sample count, and anomalous flag per model/metric pair.

  5. Review Latency percentiles — real p50/p95/p99/avg/request-count per model, sourced from a platform read. It shows “Temporarily unavailable” if that read can’t be reached.

Terminal window
curl -X POST "https://admin.arbitex.ai/api/v1/admin/monitoring/performance/summary" \
-H "Authorization: Bearer $ADMIN_TOKEN" \
-H "Content-Type: application/json" \
-d '{}'

Latency percentiles read performance/percentiles.

Available on every plan.

Latency percentiles are a platform cross-plane read; air-gapped deployments without a platform path show “temporarily unavailable,” never fabricated latencies.