Observe

Every agent run, end to end — which agents were invoked, which tools, skills and connectors they called, and what went in and out at each step.

Runs

—

Succeeded

—

Failed

—

Cost

—

Agent
Loading runs…
No runs