Testing & running

Dry-run

A dry-run executes the entire graph exactly as a real run would, with one difference: any step with a real-world side effect (an HTTP POST/PUT/ PATCH/DELETE, a message send, a paid API call) is mocked — you see what request would have been sent and what response is assumed, without anything actually happening downstream.

Use dry-run:

• Immediately after building or editing a workflow, before saving it live.
• After changing a connector or credential, to confirm the graph still resolves correctly.

Validation before you can save

Before a workflow can be saved, it's checked for structural problems:

• Dangling edges (pointing at a node that doesn't exist).
• Unreachable nodes (no path from the trigger).
• Missing trigger.
• A condition node missing a true or false branch.

These are caught as validation errors, not silent bugs discovered on a real run.

Watching a run

Every real (non-dry-run) execution appears in Workspace → Runs with a live trace:

• One row per node, appearing the moment it starts running.
• Real duration per step.
• Click any step to see its exact input and output.
• Parallel branches (from a fan-out) are shown running concurrently, in the order they actually executed — not artificially serialized.

When something fails

By default, a failed node stops the run (unless it's marked continueOnFail). The run trace shows exactly which node failed and why, so you can ask the Agent Builder to fix that specific step rather than re-describing the whole workflow.

Human-in-the-loop runs

If a run reaches a human approval / input node, it pauses — you'll see it waiting in Workspace → Tasks until someone responds. The run resumes automatically the moment a decision is made; it isn't holding a server thread open the whole time, so approvals can safely take minutes, hours, or days.