Run your workflows the way
they should actually run

It means every workflow runs independently of every other, and every step waiting on something does not consume capacity while it waits. A workflow waiting on a third party service does not hold up another that is ready to run, and a slow step in one workflow does not slow a fast step in another. The platform handles thousands of workflows in flight at once without any single one blocking another.

The platform analyses the dependency graph of every workflow automatically and runs every step that can run in parallel, in parallel. If step A feeds into C and step B also feeds into C, then A and B run at the same time and C runs once both complete. The author does not mark steps as parallel — the analysis is automatic, based on which steps consume which outputs. Workflows finish in the time of their longest path.

Every step has a configurable retry policy with controls for maximum attempts, initial wait before the first retry, the rate the wait grows between attempts (exponential backoff), the maximum wait between attempts, and the maximum total time before giving up. Error categories can retry or fail fast independently, so a transient network error retries while a permanent validation error fails immediately. Defaults apply automatically.

Every step has a configurable timeout that bounds how long it can run before it gets cancelled. The default depends on the kind of step — fast operations get short timeouts, slow ones longer — and any step can be tuned in its panel. When a timeout fires, the step is treated as a failure and the retry policy applies, so a transient hang gets retried while a step that consistently exceeds its timeout fails fast rather than hanging the automation.

A manual intervention node pauses the workflow at a specific point and waits for a human to respond before the next step runs. The configuration specifies who is notified, what context they see, and what options they have. When the workflow reaches the node, the assigned reviewer gets a notification, the workflow pauses without consuming capacity, and the next step runs the moment the person responds. Approvals, reviews, and exceptions sit cleanly inside automated workflows.

Test mode runs the entire workflow with mock data that flows through every step exactly like a real execution, except nothing touches a real system. No email goes out, no record is created or updated, no invoice is sent, no charge is processed. Every step still produces the output a real run would, so downstream steps see realistic inputs and the test verifies the whole flow end to end. The log captures the test run like a real one.