Run, watch, and debug your routine

Make a test run

Run the routine once with a small, real input. The point of the first run is not the result; it is to check that the pipeline works end to end and is not more complex than it needs to be. Run it from the Routines panel with Run, or ask the agent to run it for you.

Watch the work

Open the Routines panel. Routines that are in progress appear under Running at the top of the list, with their current stage. Select the running routine to open its graph and check its execution: each step lights up as it starts, and the status shows whether the routine is running, retrying, waiting for input, or paused.

Click any step in the graph and choose Focus thread to open the conversation of the agent working on it. From there you can check the agent's work as it happens and follow along with what it reads, decides, and writes. When the agent is working on the web, click into the browser to see exactly what it is looking at and doing on the internet.

Notice which agent is working, what step is underway, and whether Toone is waiting for input or approval. If it is waiting on you, the thread is where you answer.

Inspect the result

When the run finishes, open the produced brief and compare it with the routine’s expected sections. Look for missing evidence, unclear assumptions, or a result saved in the wrong place. The run stays in the routine's execution history, so you can come back to it later and compare it with the next one.

Improve and repeat

Change one instruction based on what you observed, then run the same routine again. Keeping the change small makes the effect understandable. If you would rather not edit alone, open the routine and ask an agent to make the change; it edits the routine while you watch, just as it did while drafting.

You completed the operating loop

You now have a project, an agent, a repeatable routine, an observable run, and a durable output. Continue with Core concepts or browse Features.