Node Editor

Run nodes and workflows

A run turns the current node settings and connected inputs into results. You can run one node, a selection, or every generative node on the canvas.

Last reviewed August 19, 2026

The node checks the work before it starts

Each generative node validates its own settings and inputs. If a required prompt, image, frame, model price, or connection is missing, the reason appears on that node and the run does not start.

If a run does not start, use the generation troubleshooting guide to check the message before trying again.

This check happens before credits are held and before an empty result box begins generating. Fix the named field, then run the node again.

Run one node

1
Prepare the node
Choose its model and settings, then add the prompt, references, or other inputs the selected model requires.
2
Check the cost on the button
The credit amount reflects the current model, options, and number of variants. A dash means that the selected model is not currently priced and the run is blocked.
3
Press Run
The node validates the request. Empty preview boxes are filled first, and any missing preview boxes are created beside the node.

Expected result

Preview boxes appear immediately in an authorizing state. After the server accepts the run and holds the required credits, they change to generating and receive the finished results.

Run selected nodes or the complete workflow

The purple Run control in the top-right corner runs every generative node on the canvas. Open its menu to run only the nodes currently selected.

Data connections provide text, images, video, documents, or collections. Trigger connections control order. A node connected by a trigger waits until its upstream step succeeds before it starts.

A data wire does not automatically run the next node

Connect the green trigger output when one step should start another. This keeps an input connection from spending credits simply because its source changed.

Choose Run or Run Overwrite

Run

Fills empty wired previews first. If more result boxes are needed, the editor creates and wires them. Existing completed previews stay unchanged.

Run Overwrite

Reuses the wired preview boxes and replaces their current results. This keeps downstream connections attached to the same holders.

Use the small arrow beside a node's Run button to set its mode. The main button then runs the mode shown on it. Workflow runs respect each node's saved mode.

Understand queued and running states

Queued: The run has accepted this node, but it is waiting for an upstream trigger or an available execution slot.
Running: The provider request has started. Result boxes show a stage label and elapsed time instead of a guessed percentage.
Error: The node or result box explains what failed. A failed preview remains available for the next run to reuse.

Stop active runs from the top bar

While any node is queued or running, the global Run control becomes Stop All Runs. Use it to stop the active work in the project. Nodes that had not started release their held credits automatically.

Stopping applies to the active project run

The Stop label shown inside an individual node is not an individual cancellation control. Use Stop All Runs in the top bar.

Credits are held before work is charged

AgentDocks first holds the estimated credits needed for accepted steps. A completed result captures its cost. Failed, stopped, timed-out, or skipped work releases the corresponding hold.

The amount displayed in the interface is an estimate based on the current configuration. The server performs the final calculation before accepting the run.

See the current plans and credit rules for allowances, top-ups, and expiry.