Dispatching agents

Dispatching sends an issue from the dashboard to one of your machines, where its runner picks it up. The run uses the Claude Code that is installed and signed in on that machine. The dashboard is the window onto that work; it never runs agents itself.

1. 1Open the issue from Issues or the Overview.
2. 2If it is still in the pool, assign it to yourself. You dispatch only your own issues, onto your own machines.
3. 3Under Run it on a machine, pick one.

A machine takes work from the dashboard when it is linked to your account, its runner is on, and Take work sent from anywhere is on, under Work from anywhere in its This machine sheet. It clones the repository with your own git when it does not have it yet. An asleep machine queues the work and starts it when it next connects.

Where that section says that for now, work starts only in Splitterm on your machine, nothing is sent from the dashboard: open the desktop app on that machine and start the issue there. For now it always says so, because work sent from the dashboard would start while nobody may be at the machine. See Work while nobody is at the machine.

An issue that waits on another shows Blocked, names what it waits for, and cannot be dispatched until those are finished.

Until the machine picks it up, the issue says which machine it was dispatched to and that it is waiting for the machine to pick the work up. After that it shows where it runs and which column it is in there.

The project's Overview puts what needs a person first: a machine offline, checks failing, or work ready for review. See Reviewing agent work.

A machine can be awake and still pick nothing up, because its runner is off or it does not take work sent from the web. The dashboard shows it as engine paused. See The runner is off.