Many agents. One machine. No collisions.
Running five agents at once on a laptop turns it into a space heater. Cube gives them their own computer, and each one its own worktree.
The parallel-agent workflow
The most productive way to use coding agents right now is to run several at once: one fixes a bug, one writes tests, one explores a new idea. You plan, hand off, and review later.
Two things get in the way. Your laptop wasn’t built to run five agents plus their dev servers and test suites. And agents working in the same checkout overwrite each other’s changes.
How Cube handles it
- A Linux computer just for your agents, so your Mac stays fast and the agents keep running when you close it.
- A worktree for every task. Each agent works on its own branch in its own directory. Nothing collides, and you merge what works.
- One sidebar for everything. See every repo, every worktree and every running session, and jump back into any of them.
- Screens per project. Group the agents and terminals for one project on one screen, and switch between screens.
Any mix of agents
Claude Code, Codex, Gemini CLI, OpenCode, GitHub Copilot, Cursor, Cline, Goose and more can run side by side on the same machine.
Questions
How do I stop parallel agents from overwriting each other?
Give each agent its own Git worktree: a separate checkout of the same repository on its own branch. Cube creates a worktree for every task, so each agent’s changes stay separate until you merge them.
Can I mix agents, like Claude Code and Codex, on the same repo?
Yes. Each session runs in its own worktree, whichever agent it is.
Why not just use my Mac?
You can, but every agent competes with your browser, your editor and your battery, and they all stop when the lid closes. A separate machine keeps them running and keeps your laptop free.