Metrics are the view across all Runs: throughput per workflow, latency percentiles (p50/p95/p99), error rates, retry rates, and deployment-level health. They answer the “what’s happening right now at the fleet level” question that individual Runs can’t.

You’ll find Metrics under Observe → Metrics in Studio.

Metrics in AGNT5 are derived from Runs, not collected through a separate telemetry pipeline. That has two practical consequences: you don’t pay for two systems, and every aggregate is one click away from the individual Runs that produced it. An elevated error rate isn’t an abstract number — it’s a filter into the Runs that failed.

A deeper guide is in progress. For broader context on production operations, see the Run overview.