Skip to main content
Observability holds the runs of your organization in four views of the same query. Its sidebar holds the view picker, your saved searches and the filters.
Observability: the query bar, runs started over time, the list of runs, and the filtersObservability: the query bar, runs started over time, the list of runs, and the filters

The query bar

Type a query: terms separated by spaces, all of which must hold. The bar suggests fields and the values that exist as you type, and underlines a term the server cannot read.
The query bar suggesting @agent:support, @agent:triage and @agent:digestThe query bar suggesting @agent:support, @agent:triage and @agent:digest

Suggestions drawn from the values that exist: here, the values of the label agent

Save keeps the query as a saved search, in your browser; Find (F) opens it later.

Filters

The sidebar lists, for each field, the values found with their counts: the timeline, the status, the namespaces, each label, the tools, the errors, the scripts. A filter is a term of the query:
  • one click asks for the value (status:failed);
  • a second click excludes it (-status:failed);
  • a third removes it.
Observability filtered on the failed runs of the triage agentObservability filtered on the failed runs of the triage agent

@agent:triage status:failed: the filters checked in the sidebar are the terms of the query

The four views

All runs

The runs started over time, then the list: each run with its status, namespace, the first line of its script, when it started, its duration, its calls, when it wakes and its error. Sort it by start, by last activity, or by next wake.

Active

The runs not yet ended.

Waiting

The active runs waiting on the outside world, grouped by what they wait for, by tab: signals, approvals, timers and tools. Each run shows the line it is paused on and when it wakes, the soonest first. Use it to find what a person, a tool or your harness has to answer.
Waiting runs grouped by signal, timer, host tool, approval and tool callWaiting runs grouped by signal, timer, host tool, approval and tool call

Failures

The failed, cancelled and aborted runs of a window, grouped by fingerprint: the error with its numbers, ids, times and quoted strings replaced, so the same failure across the many scripts models write is one group. Each group shows its runs over time, how many scripts hit it, when it was first and last seen, the line most of its runs ended at, the top labels, and the latest run.
Failures grouped by fingerprint, with their series, scripts, line and latest runFailures grouped by fingerprint, with their series, scripts, line and latest run
Click a run, anywhere, to open it in the Run view.