Launch or Run?

There are three ways to start work in ClusterCode, and picking the wrong one is the single most common source of confusion. The difference is not what they create — all three end up with a DevBox — it is who starts the agent.

Create DevBox gives you a machine. Start Agent Run gives you a result. Create Schedule gives you recurring results.

The three workflows

Workflow	What it creates	Does an agent start automatically?	Who drives	Where to watch	Best for
Create DevBox	An isolated prepared environment: repo, branch, tools, optional ticket context	No	You, from the terminal	The DevBox terminal / desktop	Hands-on exploration, debugging, pairing
Start Agent Run	One delegated AI task, in a new or existing DevBox	Yes	ClusterCode, after you submit	The Run activity stream	One ticket, completed autonomously
Create Schedule	Agent Runs triggered at a time or cadence	Yes, when triggered	ClusterCode	Schedule history, plus each Run	Recurring, nightly, batch work

Launching a DevBox does not start an AI agent.

This is the one that surprises people. A launch gives you a prepared workspace — the repo cloned, the branch checked out, the ticket context written, the agent CLI installed. Nothing is running until you start it, or until a Run or Schedule does.

Which one do I want?

You want	Use
One task, and you want to watch and steer it	Run (interactive)
One task, fire and forget	Run (autonomous)
The same task, on a cadence	Schedule
To iterate until tests or checks pass	Loop
Several engines competing, then pick the winner	Prototype race
Several engines cooperating in one job	Multi-agent run
Just a machine to work on yourself	Launch a DevBox

They compose

These are not three separate products. A Run creates a DevBox to work in, and a Schedule creates Runs. So the honest way to read the table above is: you are always getting a DevBox, and you are choosing how much of the work ClusterCode does before handing it to you.

That also means you can move between them. Launch a DevBox to look around, then hand that same prepared workspace to a Run without re-cloning anything — the launch screen’s Delegate with Agent Run action does exactly this, reusing the DevBox and carrying its ticket across.

What runs where

All three run on machines you register and control — your repositories, your credentials, your agent output, your audit trail. That is true whether the machine is a laptop in your office, a VPS you rent, or a cloud worker ClusterCode manages for you. You own the machines; you control the automation; you keep the data.

Next steps

• Run an agent on demand — the delegated path, with copy-paste recipes
• Launch from a ticket — the hands-on path, end to end
• Automate recurring work — turning a Run into a Schedule
• How a target is chosen — which machine your work lands on