Terminal-native · multi-agent
Terminal-based multi-agent orchestration with configurable external tooling, vim-like keyboard navigation, and powerful workflow patterns.
Quickstart
No daemon, no config server. Build the binary, put it on your PATH, point it at your repos, and launch. Spin up isolated worktrees in the app and start orchestrating.
Compile the release binary. One static Rust executable, no runtime dependencies.
Put the release executable on your PATH so you can call wsx from anywhere.
Register any git repository. Then create isolated worktrees for it right inside the app — one per agent session.
Open the TUI. Spin up agent sessions, review, and orchestrate — all from the keyboard.
Capabilities
Every session is an isolated worktree. workspacex gives you the connective tissue — communication, review, git, processes, and shared workspaces you can reattach to from anywhere — to drive them in parallel.
workspacex never wraps an agent CLI or calls model APIs itself. It loads each agent's own terminal UI in a PTY — Claude Code, Codex, and friends run exactly as they would in your terminal, on the subscription you already pay for. workspacex adds zero token usage and never incurs extra API charges of its own.
A workspace isn't locked to one tool. Drop Claude, Codex, Pi, and Hermes onto the same worktree from the agent picker and let them collaborate — one agent can review, hand a fix to another over the CLI, and verify the result, no matter which harness each one runs.
A built-in CLI lets agents talk to each other inside a workspace — hand off subtasks, ask questions, and coordinate without leaving the session.
The wsx and agent-review skills install during setup. Pin /agent-review to a workspace and kick off a max-effort review in a fresh session with the agent of your choice.
Original prompt, latest agent updates, git diffs, and workspace details — all in one scannable view.
Diff counts read straight from local git, and PR status shows up automatically once the GitHub CLI (gh) is installed — both reflected on the dashboard and inside agent chats.
Jump a workspace into your editor, a diff viewer, lazygit, and more — without lifting your hands from the keyboard.
Bring multiple workspaces into a single recursive pane tree and watch your agents work side by side. Split the attached view vim-style — v for panes side-by-side, s for stacked — then nest further however you like. Save any arrangement to its anchor workspace and it restores the moment you enter it, even across restarts.
Mark any workspace shared and its agents run inside tmux sessions instead of the wsx process tree. Quit wsx or lose your ssh connection — the agent keeps working, and reattaching picks the conversation up exactly where you left it. Press H to browse shared workspaces on any configured host — branches colorized by PR status — and attach over plain ssh, or skip wsx entirely with tmux attach. No daemon, no port forwarding.
View and manage every workspace process. Kick off a dev server straight from the dashboard or the agent chat view, and watch it from the same pane.
Save remote commands to launch tmux-connected sessions instantly, and pair with Tailscale to reach your full dashboard and workspaces from anywhere.
In motion
Two short walkthroughs of real workspacex sessions — heterogeneous agents and parallel worktrees.
Screencast 01
Spin up a workspace, then add any agent harness alongside the first — Claude, Codex, Pi, or Hermes — all on the same worktree. Here Claude reviews the code, hands the fix to its Codex teammate over the wsx agent send CLI, and Codex commits and reports back. Two different agents, one workspace, no human in the loop.
Screencast 02
Create isolated worktrees right from the dashboard, deploy an agent to fix and commit in each, and watch every workspace's edits and ± line counts update live in the detail bars.
Open source · MIT
workspacex is free and open source. Clone it, build the binary, and put a fleet of agents to work in five minutes — then fork it, file an issue, or send a PR.