Finding runs

The run explorer finds any run in seconds. Open it from Runs, explorer.

Filters

Stack as many as you like; results narrow as you go:

  • Agent and status (success, failed, timeout, running).
  • Evaluation state: passed, failed, pending, or not evaluated, plus a minimum quality.
  • Time window: the last hour, day, week, or 30 days. Every view is bounded by a window so results come back fast.
  • Duration and cost: minimum duration, maximum cost.
  • Metadata: any key = value pair you sent on the run. Add several to AND them together. The key box suggests keys your account has actually used.

The query bar

If you'd rather type, the bar at the top accepts shorthand that compiles to the same filters:

status:failed agent:invoicer env:prod

Reserved words (status, agent, eval) map to those filters; anything else, like env:prod, is treated as a metadata pair. Quote values with spaces: region:"eu west".

Every filtered view lives in the URL. Copy the address bar and the person you send it to lands on exactly the same filtered set. Incident links and status pages use the same pattern.

Saved filters

Name a view and it becomes a tab above the explorer, per user. Good for the searches you run every day ("failed prod runs", "expensive runs this week").

Export

Export these results hands your current agents and time window to the bulk export, so you can pull the matching runs out as NDJSON and CSV.