A workflow is a graph: a trigger starts it and nodes follow the arrows. Order comes from the arrows, not from where nodes sit on the canvas.
Triggers
- Event: any of the platform's triggerable events (226 of them, grouped by plugin).
- Schedule: a five-field cron checked every minute against the server's local time; names (MON) and 7 for Sunday are not supported, and day-of-month and day-of-week must both match.
- Manual: Run Now in the editor, with no data.
Only workflows whose status is Active run. Draft, Paused and Archived never do.
Nodes
Condition (true / false), Delay (durable), Wait for event, Action, Parallel and Join. Actions include webhook, notify, chat message, HTTP request, email, SMS, transform, log, AI prompt, and requests to other plugins (create invoice, send estimate, create booking, announce in a community, enroll in a course, update a CRM lead's status, tag a contact, update a contact field, create a task).
Requests to other plugins are fire-and-forget
Those actions emit a request and report success at once; whether the other plugin did the work is not reported. Several (invoice, estimate, booking, enrollment, announcement) carry a key built from the first run's id, so a rerun does not create a second record. CRM actions and email, SMS and webhooks have no such key and repeat.
A run pins its definition
Each run stores the graph, triggers and settings it started with. Edits apply to new runs. A rerun or resume reuses the original snapshot, so fixing a workflow and pressing Re-run runs the old, broken version; start a new run instead.
Loop guard and rate control
The loop guard allows 3 runs an hour per workflow per source (the entity id in the event, such as the order or ticket; a schedule tick counts as its own source). Beyond it the run is dropped without a history row or an alert. Rate control (max concurrent, throttle, debounce) is off by default; its skipped runs are recorded as one Skipped row a minute.
Failures
A failing action may retry (up to 5 attempts, off by default). An action node can have an on error branch; a failure down it leaves the run Completed with no alert and no dead letter. Otherwise the run fails, a dead letter is counted on Monitoring (it cannot be viewed or closed), and the failure alerts go out.
Who can do what
Everything needs the view permission; creating, editing and running needs write plus a plan with workflows; deleting needs delete. Pages show write buttons to read-only roles.