Run History
Inspect ActionFlow run states, filtered persisted results, settled credit usage, reruns, cancellation, and HITL continuations without assuming history is a raw request archive.
Run History
An admitted ActionFlow run is recorded in run history. Open ActionFlow → Run History or the corresponding panel in Actionflow Studio. A request rejected before enqueue, such as an API validation error or a webhook 422, may not have a history row. Scheduled runs can create a history row without the same pre-enqueue validation, while queue-dispatched runs can create one before queue-time validation; a row is not proof that validateFlow passed.
What is recorded
A history row can include:
- run and flow identifiers, run number, trigger source, and queue ID;
- created, started, finished, and duration timestamps;
- the product status and failure reason;
- executed node IDs, selected run metadata, and JSON-safe results;
- settled credit usage and retry/continuation links;
- generated artifact references tied to the run.
The runtime JSON can contain run inputs and trigger context, but the app's history read path sanitizes the data before showing it. It exposes selected metadata and results, filters some provider shapes, and redacts sensitive-looking keys. History is therefore a filtered audit view, not a promise of an unlimited, byte-for-byte copy of every raw input, provider response, or secret.
Do not use Run History as a secret store or as a complete request archive. Use the live Actionflow Studio output and targeted, redacted logs when you need to inspect a value immediately.
Statuses
| Status | Meaning |
|---|---|
waiting | The run is waiting in the flow queue. |
dispatched | The run was handed to the execution service. The public API may display this as QUEUED. |
executing | The graph is executing nodes. |
waitingForHitl | A HITL node paused the prefix run for a human decision. |
completed | The graph finished successfully. |
failed | The graph or admitted run failed. |
cancelled | A queued or active run was cancelled. |
unknown | The stored status is missing or not recognized. |
Execution-service terminal aliases such as COMPLETED_SUCCESSFULLY, CANCELED, CRASHED, TIMED_OUT, and SYSTEM_FAILURE are normalized for the app/API. See Execution model for the queue-to-runtime lifecycle.
Result visibility and truncation
The detail panel uses the persisted result payload after a run is terminal. Some values are transformed or omitted before display:
- credential-shaped keys are removed from the history view;
- step-shaped provider results may be reduced to their response output;
- generated media may be represented by a durable storage key and a temporary access URL;
- a loop-body result retained for another iteration is capped at 100,000 serialized characters and marked as truncated when it exceeds that cap.
That 100,000-character cap applies to the loop-history path; it is not a universal limit on every node result. A result can also be empty because the provider returned no content, the mapping resolved to undefined, or the history sanitizer filtered it. Inspect the live output before debugging a mapping as if data were lost.
Filtering
The Actionflow Studio history supports status, trigger source, and date filters. The public flow-runs endpoint returns a summary list with cursor pagination; its current status and date filters are documented by the List Runs API. Queue state is separate from run history and is exposed through the Queue API.
Actions
Rerun
The Actionflow Studio marks completed, failed, and cancelled rows as rerunnable through canRerun. An Actionflow Studio rerun:
- validates the current saved graph;
- creates a new run linked by
retriedFromRunHistoryId; - uses the manual trigger source and can consume credits again.
Current limitation: Rerun from Actionflow Studio cannot reliably restore prior run-scoped inputs. The Actionflow Studio history allowlist omits inputs and metadata.inputs, so its rerun action has no previous input map to pass. To reproduce a run with the same inputs, use the original trigger/request or another supported rerun surface/API behavior that preserves the input contract. Do not rely on Actionflow Studio Rerun to restore prior inputs.
A rerun starts the current graph from its beginning. It is not an in-place checkpoint resume, and it can repeat external side effects such as messages, webhooks, or writes. Do not use Rerun to recover a HITL continuation; use the HITL decision flow.
Cancellation
A waiting queue item with a queueId can be cancelled from Actionflow Studio. An active Actionflow Studio run can be stopped while it is in a cancellable execution state. Completed, failed, or already-dispatched rows are not guaranteed to have a cancellation action. After cancelling, reload the row and confirm its terminal status and credit record.
HITL continuation
When a HITL node pauses a run, the prefix row is stored as waitingForHitl with the partial execution state. A valid human decision attempts to queue a new continuation run from the HITL node, restores the saved results and Loop context, and links the continuation to the prefix. The prefix row is marked completed only after the continuation is enqueued; if enqueue fails, the callback can still return 202 with resolved: false. The two rows should not be treated as one continuous row in history.