Skip to main content

Search across runs

Find a run by any text it touched — step outputs, results, errors, and trigger payloads, searched across every run in the workspace.

Sometimes you know what happened but not which run it happened in: a customer quotes an invoice number, a teammate forwards an error message, a webhook carried an order id. Search answers that question directly. It matches the text that moved through your runs — what each step produced, what each run returned or failed with, and the payload that started it — across every run in the workspace, and takes you to the run that carried it.

When to use this

Reach for search when the run id is what you are missing. If you already have the run id or you are working through one run's timeline, open the run instead — the run detail page shows the full step sequence with per-step logs, gates, and retries.

Search from anywhere in Monitoring

The search box sits in the Monitoring header, above Activity, Runs, and Approvals. Click Search runs, or press ⌘K (Ctrl+K on Windows and Linux) from any Monitoring page.

  1. Type at least two characters

    Results start after two characters and refresh as you type. Queries are capped at 200 characters — a phrase works better than a paragraph.

  2. Read the grouped results

    Matches are grouped by run, newest and most relevant first. Each match is labeled by what it is — a step output, a step error, the run's result or error, or the trigger payload — with a snippet highlighting the matched words. Errors and results rank above step-level matches for the same words.

  3. Open the run

    Click a result — or highlight it with the arrow keys and press Enter — to open that run's detail page. A step-level match lands on the step that produced the text.

What search can and cannot see

Search reads the same redacted record the run detail page shows you. Values your workspace's redaction rules mask are masked before they are ever indexed, so search can never surface something the run page would hide. It is scoped to the workspace you are in: it never returns a run from another workspace or another account.

Search covers five kinds of text from every run, successful or failed:

  • Step outputs — what each step produced.
  • Step errors — the message and code of a failing step.
  • Run results — the final output of a successful run.
  • Run errors — the reason a run failed. Failed runs are found by the text of their failure.
  • Trigger payloads — the body that started a successful run.

Three bounds worth knowing:

  • Trigger payloads of failed runs are not searchable. A run that fails never records its starting payload for search; its error text is searchable instead.
  • Step, error, and trigger history begins when expanded coverage reached your account. Earlier runs remain searchable by their results only.
  • A step that ran in a loop is searchable by its latest iteration's output. Search finds the step; the run page shows every iteration.

Search covers text, not structure. It matches words in the indexed record, so invoice 4021 finds a run that carried that invoice number. It cannot answer "which runs failed on Tuesday" — that is what the run list filters and the Activity feed are for.

Plan availability and the account toggle

Run search is available on the Scale plan. Account owners can turn indexing off for the whole account under Settings → General → Run search. Turning it off permanently deletes the account's search index; if it is turned back on later, coverage begins again from that moment — earlier runs are not re-indexed.

When a matched run has aged out

Runs expire on your plan's retention window (see Data handling). A search hit whose run has already expired still shows you the snippet — that text is the answer you were looking for — but it has no link, and the run id is marked as expired. Copy what you need from the snippet; there is no page left to open.

Search returns nothing

No matches at all. Search matches whole words, not fragments: 4021 will not match INV-4021-B as a substring. Try the full token as it appears in the run. Widen to a distinctive word from the message rather than an id you are guessing at.

The text predates coverage. Step outputs, errors, and trigger payloads are indexed from the moment expanded coverage reached your account; older runs are searchable by their results only, and runs that completed before search became available in your workspace are not indexed retroactively.

Run search is off for the account. An account owner disabled it under Settings → General → Run search; indexing resumes from the moment it is re-enabled.

The run is older than your retention window. Expired runs leave the index with everything else. Confirm the timeframe against your plan's window in Data handling.

"Search timed out." A very broad query over a busy workspace can exceed the time budget. Add a second word to narrow it and run the search again from the same box.

Was this helpful?
Replay a runWhat your clients see in the portal