Getting Started

Quickstart

You are absolutely right — at scale.

Running several coding agents in parallel means tabbing between terminals, tracking which agent is doing what, or running specialized tools that don't fit your workflow.

actor.sh manages that coordination for you — you describe the work in your main session and parallel agents carry it out in the background. Each one gets a full claude or codex session, by default in a git worktree. Press Ctrl-Z at any time to switch between them, watch the work happen, or chat with it directly.

Run it

First, install actor.sh:

curl -fsSL https://actor.sh/install.sh | sh

(To build from source instead, see Install from Source.)

Then launch the main claude session:

actor claude

This is just regular Claude Code, but notice there is now a sticky menu at the top of the screen, collapsed into a small pill:

actor.sh menu

The pill shows where you are — main. (You can press Ctrl-Z to expand it into the full menu — see the session menu.)

Ask for what you want done:

Spawn a session to add rate limiting to the public API endpoints.

actor.sh will spin up a new session — picking a name like rate-limit — hand it the task, and run it in the background. By default it gets its own branch and worktree at ~/.actor/worktrees/rate-limit/; your checkout is untouched. When it finishes a turn, the result reaches main over a notification channel. main verifies the work, surfaces what needs your attention, and otherwise stays quiet.

Clean up

When you're done with a session, ask main to discard it. This removes its worktree and its saved definition:

We're done with rate-limit — discard it.

If the worktree has uncommitted changes, the discard is refused — actor.sh won't silently drop work. Commit (or push) first, or ask for a forced discard to override. The branch itself is never deleted.

Where to go next

  • Session Menu — the menu in full: every session, attach, scrollback.
  • Remote Hosts — run sessions on other machines.