Muxie Get Muxie
Getting started Tutorial

Running a coding agent

Start a coding agent in a pane, watch Muxie label it working, blocked, or done, and reattach with full context after a disconnect.

Muxie is built around coding agents. This walkthrough starts one, shows how Muxie tracks its state, and proves the session survives a disconnect.

We’ll use Claude Code as the example, but the same applies to Codex, Aider, Gemini, Cursor Agent, OpenCode, Amp, and Droid — see Supported agents.

1. Start the agent

Open a terminal pane (Ctrl+T → Terminal), change into your project, and run your agent as you normally would:

cd ~/code/my-project
claude

Muxie detects the agent automatically from the process running in the pane.

2. Watch its status

As the agent runs, its session card in the left rail shows a status chip:

  • Working — the agent is busy.
  • Blocked — the agent is waiting for you (a permission prompt, a question).
  • Done / idle — the agent finished or returned to a shell prompt.

The card also shows a short recap of the most recent activity, so you can glance at a quiet pane and know where it left off. You can hide or show these chips under Settings → Sidebar.

Tip: For the most accurate status, install your agent’s lifecycle hooks so it reports state changes directly instead of relying on screen detection. See Install agent hooks. Muxie can install Claude Code’s hooks for you with muxie hooks setup claude.

3. Use the Agents hub

Open the Agents hub (a panel on the right; also reachable from the command palette). It lists every agent across all your workspaces — local, SSH, and sprites.dev — grouped by workspace and split into Working, Blocked, Done, and Idle. From a row you can jump straight to the agent’s pane.

This is the calm overview: when several agents are running, the hub tells you which one needs attention without hunting through panes.

4. Get notified

When an agent becomes blocked or finishes, Muxie can play a sound and surface a notification — useful when you’ve tabbed away. Configure this under Settings → Notifications, including a custom sound, a shell command to run on each notification, and mute patterns. See Set up notifications.

5. Survive a disconnect

Leave the agent running and close the window — or, for a remote agent, drop the network. The agent keeps running inside the daemon (locally) or on the host (SSH/sprites.dev). Reopen Muxie and the pane reattaches to the same session with its scrollback, so the agent’s context is exactly as you left it. No restart, no re-priming.

Spawning agents from the CLI

Inside a Muxie pane you can script new agent panes with the muxie CLI:

# open a new pane to the right running an agent with an initial prompt
muxie spawn right claude --prompt "review the staging deploy"

muxie spawn prints the new session id, and there are blocking helpers like muxie api wait.agentState --state blocked to orchestrate multi-agent flows. See the CLI reference.

Next

Understand what’s happening under the hood in Agent awareness, or learn why the session survived in Durable sessions.