Hang4r: Run Claude Code, Codex, Cursor in Parallel

A new macOS app called Hang4r lets you run Claude Code, OpenAI Codex, and Cursor coding agents side by side, up to four at once, each locked into its own isolated copy of your repository. It is free for Apple Silicon Macs, in early access, and it wraps the command-line agents you already pay for rather than asking for a new API key. The pitch is simple: stop babysitting one agent at a time and start dispatching several in parallel, then review what each one did from a single window.

That idea is not new, but the field around it has gotten crowded fast, and the tools differ in ways that matter for how you actually work. This guide breaks down what Hang4r does, why parallel agents need repository isolation to be safe, how it stacks up against Conductor, Parallel Code, and cmux, and a concrete workflow for running three agents against one codebase without them stepping on each other.

What Hang4r Actually Does

Hang4r is a desktop control surface for coding-agent CLIs. Instead of opening three terminal tabs and juggling three sessions, you get a tiled workspace where each pane hosts a full agent session: a real PTY terminal, a built-in Monaco code editor, and an embedded browser. It drives the CLIs you already have installed, so your existing Claude Code subagents, hooks, skills, and MCP configuration carry over untouched. The same holds for the OpenAI Codex CLI and the Cursor agent.

The features that separate it from a plain terminal multiplexer are the ones built around review and safety. Every session runs in its own git worktree, so two agents editing the same file never collide. Hang4r commits a checkpoint after every agent turn, which means you can rewind a session to any point instead of losing work to an overeager rewrite. A diff review panel lets you leave inline comments that get fed back to the agent as follow-up instructions, closing the loop between reading the change and asking for a fix. It also tracks token cost and rate-limit status per account, imports your existing CLI session history, and exposes a hooks timeline and subagent inspector so you can see what the agent is actually doing.

Why Parallel Agents Need Isolation

The reason every serious tool in this space leans on git worktrees is that running multiple agents on a single checkout is a recipe for chaos. Branch switches interrupt a running agent. Stashing kills its momentum. And two agents trying to edit the same file at the same time produce merge conflicts that take longer to untangle than the original task would have.

A git worktree solves this by giving each agent its own directory on its own branch, checked out from the same repository. The worktrees share history, objects, and refs, so a commit made in one is immediately visible from the others, and you never duplicate the .git directory. Claude Code has this built in: passing a worktree flag spins up an isolated directory automatically, and you can make isolation permanent for a subagent, as the Claude Code worktree docs describe. Hang4r automates that setup so you do not have to create, name, and clean up worktrees by hand. Most developers run two to five agents in parallel this way without any dedicated infrastructure.

Hang4r vs the Parallel-Agent Field

Hang4r arrives into a category that already includes a native Mac incumbent and two open-source contenders. Conductor is the most direct comparison: a Mac app that runs Claude Code, Codex, and Cursor in isolated workspaces with a unified review-and-merge interface. On the open-source side, Parallel Code is MIT-licensed, runs on macOS and Linux, and dispatches Claude Code, Codex, Gemini, and Copilot CLIs each in its own worktree, while cmux casts the widest net, running Claude Code, Codex, Amp, Gemini, Cursor, and Opencode across many tasks at once. The table below lays out the practical differences.

Hang4r's differentiators inside that group are the per-turn checkpoint commits and the inline-comment diff loop. Conductor covers the same three commercial agents but does not advertise turn-level rewind. If you want the broadest agent coverage or a license you can audit, the open-source options win; if you want a polished Mac app with fine-grained undo, Hang4r and Conductor are the contenders.

How to Run Three Agents in Parallel Without Collisions

You do not need Hang4r to try the parallel-agent workflow, though it removes the manual steps. Here is the pattern, which you can run today with plain Git and any coding-agent CLI.

1. Split the work into independent tasks. Parallelism only pays off when tasks do not touch the same files. Pick, for example, a bug fix in the auth module, a new API endpoint, and a docs update. Overlapping tasks will still collide at merge time.
2. Give each task its own worktree. Run git worktree add ../feature-auth -b feature-auth for each task, creating a separate directory on a fresh branch from main. In Hang4r or Conductor this happens automatically when you open a new session.
3. Launch one agent per worktree. Point Claude Code at the first directory, Codex at the second, Cursor at the third. Each agent sees only its own working tree and cannot disturb the others.
4. Review before merging. When an agent finishes, read its diff. In Hang4r you leave inline comments that go straight back to the agent; with raw Git you run git diff main and iterate in the terminal.
5. Merge in sequence, resolve once. Bring the branches back to main one at a time. Because the tasks were scoped to different files, conflicts should be rare, and you handle any that appear a single time rather than mid-run.

What This Enables

The payoff is throughput. A single agent forces you into a wait-then-review rhythm where you sit idle while it thinks. Three agents in worktrees turn that dead time into parallel progress: while one is generating, you are reviewing another and steering a third. For a solo builder shipping features, or a small team spreading tasks across models to see which agent handles which job best, that is the difference between one thing at a time and a small fleet. Hang4r's contribution is making the isolation, checkpointing, and review invisible, so the overhead of running four agents feels closer to running one.

What to Do Next

If you are on an Apple Silicon Mac and already pay for at least one of Claude Code, Codex, or Cursor, install Hang4r and start with two agents on two genuinely independent tasks before scaling to four. If you are on Linux, or you want an open-source tool you can inspect, start with Parallel Code or cmux instead. And if you would rather understand the mechanics first, spend ten minutes running the manual git worktree workflow above; it is the foundation every one of these apps is built on.

Frequently asked questions

Is Hang4r free?

Yes. Hang4r is free during its early-access period for Apple Silicon Macs. It does not require a separate API key or subscription of its own, but it drives coding-agent CLIs like Claude Code, Codex, and Cursor, so you still need an active subscription to at least one of those.

What is the difference between Hang4r and Conductor?

Both are Mac apps that run Claude Code, Codex, and Cursor in parallel with isolated workspaces and a unified review interface. Hang4r's distinguishing features are per-turn checkpoint commits that let you rewind any session and a diff panel where inline comments are fed back to the agent as follow-up instructions.

Do I need to know git worktrees to use it?

No. Hang4r creates and manages a git worktree for each session automatically. Understanding worktrees helps you reason about what the tool is doing, but you never have to run the commands yourself. The same worktree isolation is available directly in Claude Code if you prefer the command line.

Can I run agents from different providers at the same time?

Yes. That is the core use case. You can run a Claude Code session, a Codex session, and a Cursor session simultaneously, each in its own worktree, and compare how the different models handle their assigned tasks.

Does it work on Windows or Linux?

Hang4r is macOS only and currently targets Apple Silicon. If you need Linux support, Parallel Code runs on macOS and Linux, and cmux is cross-platform. Both are open source and free.

How many agents can I run at once?

Hang4r tiles up to four sessions in its workspace. In practice, most developers find two to five parallel agents is the sweet spot before review overhead outweighs the throughput gain, regardless of which tool they use.