Skip to main content
client.executions is a read-only view of what ran: workflow runs, table-automation runs and app business-rule runs. There is no rerun on this surface. It wraps /v1/executions* - see the API reference.
executions.data returns recorded payloads, and payloads can hold customer data - whatever the step sent and received. Use a least-privilege key for anything that reads step data, and do not log what it returns.Tool and MCP runs are readable by any key in the same workspace and environment. A key sees the executions of its own workspace and environment, whoever started them, including runs started through a tool call or an MCP server. This is accepted behaviour, not an oversight. An account-wide key (created before keys were tied to a workspace) sees the account’s executions. See API keys.

List executions

Filters (Python snake_case, TypeScript camelCase): With no created_after / created_before, the API limits the query to the last 30 days.

Paging

list without top drains every match: it follows the cursor the server returns until none is left. Passing top returns that one page (server cap 100); add cursor (a previous page’s next_cursor / nextCursor, from list_page / listPage) or offset to read a later one. offset and cursor without top raise EnginiClientError, because without top every page is drained from the start.
Cursor paging works only for a single createdAt or executionId sort key. Sorting by closedAt, or by several keys, has no cursor on the server, so list drains by offset instead. Rows that arrive mid-drain shift the pages: the SDK de-dupes by executionId but can still skip a row, so re-run with a fixed created_before for a stable window.

Get one execution, its log and step data

  • get also serves to follow an asynchronous run: poll it until status is terminal. duration is omitted while the run is open.
  • log returns one entry per designer step, not per loop iteration - a step inside a 5,000-row loop is one entry. Use data for per-iteration detail.
  • A log entry carries error_message (errorMessage in TypeScript) only when its status is ranWithError: the failed step’s text, cut at 300 characters (a trigger that stops on its condition reports Fell By Condition). For a connector call the full error body is in data(..., type="response"). The text can come from a third party, so treat it as untrusted data.
  • data(execution_id, step_no, ...) takes the step number from log. It narrows with type (request, response, input or status), loop_path / loopPath (a list of step:iteration, outermost first), run_number / runNumber, retry_no / retryNo and status. Without top it drains every page (the server returns at most 10 per page); with top you get one page, and offset is valid only with top.
  • A payload that could not be read from storage comes back with value_unavailable / additional_unavailable (valueUnavailable / additionalUnavailable) set, rather than failing the call.

Retention

log and data raise EnginiAPIError (HTTP 410) once the execution’s per-step detail has passed its workflow’s history retention window. The execution itself stays listed, with data_available / dataAvailable false.