zeus

I was running Claude Code, Codex, Cursor, and a pile of shells, and keeping track of them meant scanning terminals. Which agent was still working? Which one needed something from me? Closing a window could also mean killing the work inside it.

Zeus grew out of that. I built a native macOS workbench in Rust with GPUI, with a real terminal for each agent and a sidebar that reduces its state to working, waiting, or done. I still have the terminals when I need the details. I don't have to read all of them just to find out who is waiting.

Closing the window shouldn't stop the work

The main architectural decision was to keep the processes out of the UI. A background Engine owns the sessions, Git worktrees, and SSH remotes. The window is a way to reach that work, not the thing keeping it alive.

That separation takes more machinery than putting terminals inside an app: the Engine, PTY holders, and a versioned protocol between the pieces. But I can quit and reopen Zeus without ending the agent sessions.

More than one way in

I made agent support catalog data rather than hard-coding each CLI into the app. Adding another agent is relatively cheap that way, though detecting what it is doing still relies on heuristics.

Through MCP, a running agent can spawn and watch another agent. I also built a SwiftUI iOS companion that connects to a local HTTPS sidecar, so I can check the sessions from my phone. Both give me ways to reach the work without keeping the Mac window open.

Zeus is open source. The docs (opens in a new tab) cover the workbench, and the code is on GitHub (opens in a new tab).