Start an agent
Choose what the agent works on, tell it what to do, and follow along as it works.
An agent keeps its own computer and picks up where it left off. Every agent runs Claude Code.
What do I need before starting?
A repository the agent can reach. If none is listed, connect GitHub or connect GitLab first.
How do I start an agent?
- Open Agents.
- Under Works on, choose a project or up to five repositories.
- Tell the agent what to work on, then select Start agent or press Return.
The agent is named after the first line of your message. It starts its computer, clones the repositories, and gets to work. Its output streams in as it goes.
How do I give it more instructions?
Type in Tell the agent what to do next and press Return. While the agent is working, stop the current turn before you send something else.
How long can a message be?
Up to 20,000 characters.