Switchyard

English · 简体中文

Turn your own computers into an always-on AI coding fleet.

Dispatch Claude Code, Codex, or Kimi from your phone.
Let the work continue in isolated environments, then come back to review or take over.

What Switchyard is

Switchyard is a **local-first control plane for AI coding agents**. It runs on your development machine, gives every task a real git worktree and tmux session, and exposes the same working state to any browser you authorize.

• Close the browser; the task keeps running. tmux, not the page, owns the session.
• Run several tasks without collisions. Every task has its own branch, worktree, and terminal.
• Move between laptop and phone without losing context. Read progress, answer a prompt, or enter the live TUI.
• Use the agent that fits the task. Claude Code, Codex, and Kimi Code—including a per-task model such as Kimi K3—can share one board.
• Add more computers without creating a central server. Every machine remains a complete, independent Switchyard node and the sole owner of its own work.

The network layer stays deliberately thin:

phone / any browser ── private HTTPS ──► Switchyard on machine A
                                                │
                                                └── SSH control ──► Switchyard on machine B

machine A owns: A's repos · database · worktrees · tmux sessions · agents
machine B owns: B's repos · database · worktrees · tmux sessions · agents

Machine A can ask B to create or control a task, but B performs the operation and stores the result. A remote machine without Switchyard is n