Docs / Agents

Local commands and terminals

Let an agent run one-off commands on a selected computer, or share an interactive terminal with it.

Select the computer a conversation may use before you ask the agent to run work there. The agent can use that computer's CLIs, files and sign-ins, so name the device and folder when they matter. A computer being online doesn't prove it is the right device or the right cloud account.

A command that finishes

To inspect files, run git, install dependencies or run a build, the agent uses local_exec. It returns the output and an exit status. Commands that run too long may be stopped. Use a terminal for processes meant to stay open.

You can rename how this Mac appears in the computer picker under Settings → This computer → Local exec.

A shared terminal

Ask for a terminal when you want to see and interact with a shell, a development server, a watcher, a log tail or a CLI waiting for input. local_terminal opens it in the dock. You and the agent can both type into it. Stop the foreground process or close the terminal when you're done. Closing the tab ends its shell.

Terminal output is a recent snapshot, not a finished command's exit status. A running terminal is not a durable schedule or monitor. Use triggers for future work that must be saved.

Approval and access

Local actions follow the conversation's permission mode. In Auto Mode, the runtime decides which actions need approval. Full Access approves permission requests automatically. local_exec has no extra "ask before every command" step of its own.

Local commands run with the computer user's permissions, including any cloud and repository sign-ins already there. Check that identity before cloud administration. The credential picker for connected tools does not remove logins already on a local machine. Read Runtimes and computers before you share a computer.