Running agents

Running an agent executes its graph and produces results on each node. This guide covers how a run flows, how to read results, how to batch over many inputs with loops, and how to pause a run for approval.

​

Triggering a run

You start a run from the studio’s Run button. Pleyor first saves the current graph, then executes it. You can also re-run a single node when you only want to iterate on one step instead of the whole pipeline. Each run captures the current input values — your text prompts and uploaded files — at the moment you trigger it. Edits you make to inputs while a run is in flight apply to your next run, not the one already executing.

​

Per-node execution

A run walks the nodes in order. For each node, Pleyor:

1. Decides whether the node should run or be skipped. A node is skipped when a required input is missing, an upstream branch wasn’t taken, or its inputs haven’t changed since the last run.
2. Runs the node if it should execute, calling the underlying model or transform.
3. Records the output and reports the result back to the studio.

You watch this happen live. Nodes flip to the generating state as they start and to completed (or error, or skipped) as they finish. No refresh needed — the canvas updates as the run progresses.

​

Viewing outputs

When a node completes, its output renders on the node — an image, a video player, an audio player, or text. Click any node to inspect what it produced and the inputs it received.

​

Loop iterations

Some nodes produce more than one result, and downstream nodes run once per result. This is how you batch: feed a generator multiple variations or a list of inputs, and each downstream node executes once per item as a separate iteration. Each iteration is tracked independently. A node that ran over several inputs shows each iteration’s result, so you can compare variations side by side. If you reduce the number of items between runs, the leftover iterations from the previous run are cleared so you only see current results.

​

Human-review pauses

A Human review node pauses the run and waits for you before continuing. When the run reaches it, the run enters a paused state and the node shows the items awaiting your decision. This is how you keep a human in the loop — nothing past the review node runs until you sign off.