Run an agent

Running a graph in Agent Playground executes every node your edges connect, and gives you a live account of what happened at each one: which node ran, what it received, and what it returned. This guide covers pressing Run Agent Workflow, reading a run while it’s in progress, and finding it again after you’ve left the builder.

Running assumes a graph already exists and is saved. This guide runs Invoice Triage, built in Build your workflow; build yours first if you haven’t. Two more things have to be true before a run starts.

• Every node on the canvas needs to be configured, or the run is refused with “Node not configured” for one node, or “3 nodes are not configured” when more than one is missing setup. Configure an LLM Prompt node or Configure an Agent node to clear this
• Every variable the graph needs has to be filled in, or the run is held and the Variables drawer opens on its own with “Fill in all variables before running”. See Set input variables

Fix whichever one is blocking you and try again.

Start the run

Open your graph in the Agent Builder tab, then press Run Agent Workflow to execute the graph. The label switches to Rerun Agent Workflow the next time you run it, so the button itself tells you whether this is a first run or a repeat.

If you’ve made changes you haven’t saved, a dialog titled “Unsaved Changes” stops you before anything runs: “You have unsaved node changes. Running now will use the last saved configuration.” Click Run Anyway to go ahead with the last saved version, or close the dialog and save first if the run needs to reflect your edits.

Watch it run

Once the run starts, a run panel opens at the bottom of the builder. It lists each node by name, and Show Outcome and Hide Outcome fold the panel away or bring it back without stopping anything underneath.

While it runs, the node currently executing is the one animating on the canvas. A node marked failed didn’t complete, and any node downstream of it is marked skipped rather than running. See Limits & rules for the full list of statuses.

Click a node inside the panel, including one marked failed, to see what went into it and what came out. Click the classifier and you’ll see the output it handed off, the same value the router receives as its input.

Once the agent finishes, the panel selects the last node that ran.

If the run as a whole can’t complete for a reason that isn’t pinned to one node, the error you see falls back to a generic “Workflow execution failed”. Check the panel’s per-node statuses for one marked failed, and click it to see what it received and returned.

Leave it running

Click Exit Workflow while a run is still going and a dialog titled “Leave running workflow?” asks first: “Your workflow will run in the background. You can find it in the Execution tab.” Choose Leave to step away, or Cancel to stay and keep watching.

Exit Workflow doesn’t stop the run. A toast confirms it: “Exited workflow. It will continue running in the background.” The run keeps executing with the builder closed.

Find it in Executions

The Executions tab sits across the top of the agent alongside Agent Builder and Changelog; open it to find that run again, or any other run of this graph. A list of runs sits on the left; select one and its per-node detail loads on the right, the same kind of detail the run panel showed while it was in progress. When a step is an Agent Node, opening it opens the nested run inside it too.

If nothing has run yet, the tab shows “No executions yet”, with “Run your workflow from the Agent Builder to see results here” underneath.

Dive deeper