Tools to run coding agents in parallel
Most tools run each agent in its own git worktree on your machine, from a terminal (Claude Squad) or a desktop app (Conductor, Crystal, Vibe Kanban). Cloud services give each task a machine of its own (Claude Code cloud sessions, Codex cloud). Murmell puts agents from several vendors in one shared cloud directory that several people can watch live.
The tools, side by side
| Tool | Where the agents run | How they are kept apart | Who can watch |
|---|---|---|---|
| Claude Squad | Your machine, in the terminal | A tmux session and a git worktree per agent | You, at that terminal |
| Conductor | Your Mac, in a desktop app | A workspace per agent, with its own branch | Whoever is at that Mac |
| Crystal | Your machine, in a desktop app | A git worktree per session | You, in the app |
| Vibe Kanban | Your machine, behind a task board | A workspace per task | Whoever is looking at the board |
| OpenHands | Your machine, or a host you run | A git worktree per agent | Whoever is at the machine |
| Claude Code cloud sessions | Virtual machines Anthropic manages | One machine and one branch per session | A shared link, showing the state when it is opened |
| Codex cloud | OpenAI’s cloud | A dedicated environment per task | The summary and the diff that come back |
| Murmell | A cloud machine per project, in your browser | One shared directory, and a file is reserved by the agent writing it | Anybody with the link, live |
Four questions that pick the tool
Do the agents need each other’s work while they run? With a worktree each, an agent building the interface cannot use the endpoint another agent wrote an hour ago until the branches are merged. In one shared directory it can use it as soon as it is saved. Git worktrees for AI coding agents goes through that trade.
Should it keep running with your laptop closed? The local tools stop when the machine sleeps. Cloud sessions, Codex cloud and Murmell keep going. Run Claude Code in the cloud compares the options, including a server of your own.
One vendor or several? Anthropic’s and OpenAI’s cloud services run their own agent. Claude Squad, Crystal and Murmell run Claude Code and Codex together.
Is anyone else watching? Local tools show the work to whoever sits at the machine. On Murmell a teammate opens the same canvas and sees every terminal as it moves. Sharing a coding session shows how.
Straight answers
What is the best tool to run several Claude Code agents in parallel?
It depends on where the work should happen. On your own machine, Claude Squad (terminal) and Conductor or Crystal (desktop) give each agent a git worktree. In the cloud with Claude alone, Anthropic's cloud sessions run each task on its own machine. For agents from several vendors and several people on one project, Murmell runs them in one shared directory.
Can I run Claude Code and Codex in parallel?
Yes. Claude Squad and Crystal run both on your machine, each in its own worktree. Murmell runs both in the cloud, side by side in the same project directory, with a reservation on each file an agent is writing.
Why do most of these tools use git worktrees?
A worktree gives each agent its own copy of the files on its own branch, so two agents never write the same file at the same moment. The price is that neither sees the other's changes until the branches are merged, and the conflicts wait for that merge.