Monitoring your agents
How to track runs, understand errors, and stay on top of your active AI agents.
Your workspace — day-to-day view
Your workspace is where you'll spend most of your time. Your agents are listed in the sidebar — each one shows a small status dot, the agent name, and when it last ran ("Run 2 hours ago").
- Emerald dot — the agents are set up and active
- Red dot — the install encountered a problem during setup
- Grey dot — the agents are still being set up
Click any agent to open its room thread. This is where you'll see recent run cards, the output from each run, and any error messages that need attention. Run cards update automatically while a run is in progress — no refresh needed.
The Observability panel
Click Observability at the bottom of the sidebar to open the Observability panel. It shows every agent design you have installed, each as a collapsible row.
Each row shows at a glance:
- Total number of runs recorded
- Overall success rate (green above 90%, amber 70–90%, red below 70%)
- Total tokens used
Expand any row to see:
- An HTTP calls summary — total external calls made, success rate, and average calls per run
- A Recent runs list (last 12 runs) — each run shows its status, the agent key or error message, token count, and timestamp. Click a failed run to expand the full error message.
Run statuses explained
Completed — The run finished successfully. The output and per-agent breakdown are available in the run card in the thread.
Failed — The run encountered an error. Open the run card in the thread or expand the run in the Observability panel to see the error message. Common causes: expired credentials, rate limits on a connected service, or a misconfiguration in the agents.
Smart alerts
Envelope watches your agents automatically and surfaces two types of alerts as inline cards in the room thread:
Run failures — Appears when the last 3 consecutive runs all failed after a period of successful runs. Something has likely changed — check credentials and any connected services.
Token spike — Appears when a run uses more than double your recent token average. This can indicate a loop, unexpected data volume, or a prompt change worth reviewing.
These alerts appear directly in the thread where they're most visible — no configuration needed.
Cost tracking
Each run records its token usage. The Observability panel shows total tokens per agent design. Pricing details (per-run fee or per-1K-tokens rate) are shown on the agent detail page before you install.
Billing note: You're billed per run at the rate set by the design author. Free agents have no run cost.