Monitor
Run history and live monitor
Every execution — agent or workflow, test or production — lands in Monitor with status, duration, cost and a one-click path into the full trace.
What you'll learn
- How to read the run list and what each column means
- How to filter and search across runs
- How to open a run and what is inside the detail view
- How to hand a failing run to Eval for scoring
The run list
Monitor opens to a grid of every recent run. New runs stream in live — no refresh needed.
- 1
Status
Success, Running, Failed, Cancelled, or Awaiting approval. Color-coded so a failing batch is obvious at a glance. - 2
Agent or workflow
The name and version that handled the run. Click through to open the source definition. - 3
Duration and timestamp
Wall-clock duration plus the start time. Sort either column to find slow runs or the most recent batch. - 4
Cost and tokens
Estimated dollar cost and total tokens consumed across LLM calls in the run.
Filtering and search
- 1
Filter by agent or workflow
Pick one or more from the dropdown to narrow the list. Combine with status to isolate failing runs of a specific agent. - 2
Filter by status
Toggle Success, Running, Failed, Cancelled, or Awaiting approval. Multi-select supported. - 3
Date range
Pick a preset (Last hour, 24h, 7d, 30d) or a custom window. The list refreshes in place. - 4
Full-text search
Search across run IDs, input payloads, and error messages. Useful for finding the run a customer ticket points at.
Opening a run
Click any row to open the detail view. You will see the original input, the final output, a chronological trace of every LLM call and tool invocation, the decision tree for agent reasoning, and the timing breakdown per step. Use the Open in Eval button to send the run into a labeling or A/B dataset.
Frequently asked questions
- How long are runs retained?
- Runs are retained for 90 days on the standard plan and 1 year on enterprise plans. Export to your own warehouse via the Runs API if you need longer retention.
- Can I share a run with a teammate?
- Yes. Every run has a permalink. Anyone in the workspace with at least Viewer access can open it.
- Does Monitor show test runs from the agent builder?
- Yes, with a Test badge to distinguish them from production traffic. Filter by source to hide test runs from the main list.
- Can I re-run a failed execution?
- Yes. The Retry action replays the run with the original input. The retried run links back to the original for audit.