Parallel Code is a macOS app that gives every AI coding agent its own git branch and worktree — automatically. Use Claude Code, Codex, and Gemini in parallel. Free and open source.
I've been maintaining an open source productivity app (Super Productivity) for 9 years and working as a programmer for much longer than that. When AI coding agents came along, they changed how I work - but the other tools and habits I grow comfortable with over the years, weren't a good fit anymore.
One agent works on a feature. I wait, get distracted, start something else. Finally, I come back, review, start the next one. Wait again. I had a backlog of ideas and the tools to build them, but things got messy. I was doing a lot, but I felt like I'd lost control compared to how I used to code.
I tried the obvious fixes. Multiple terminal windows were my new go-to solution, but got chaotic fast - which terminal was working on what? Tmux helped organize it, but I was still cycling through panes, switching tools for reviewing diffs and history, losing track of context, and dealing with merge conflicts when agents touched the same files.
So I built Parallel Code to fix this for myself.
The core idea: every task gets its own git branch and worktree automatically. Agents work in complete isolation - they can't conflict. I kick off 5, 8, 10 at a time, watch them all in a tiled view, and review diffs as they finish.
One design decision I'm particularly proud of: it runs your real terminal CLIs. Claude Code, Codex CLI, Gemini CLI - whatever you use today works exactly the same inside Parallel Code. Same behavior, same config, nothing changes. I didn't want to build another wrapper or chat UI. If you don't like the app, go back to raw terminals. Zero switching cost.
Some things I added along the way also turned out very useful:
- Phone monitoring - Scan a QR code, watch all your agent terminals from your phone. Sure there are other ways to do this, but this is really comfortable.
- Keyboard-first design - 40+ shortcuts. When you're managing 10 agents, reaching for the mouse is too slow. Best part is to navigate panes with just Alt+Arrow-keys.
- Built-in diff review and merge - The merge step is where quality happens. So having a quick way to do a code review is very useuful.
Parallel Code is free, open source (MIT), and always will be. No subscriptions, no platform fees, no telemetry. Your API keys stay on your machine.
I built this because I needed it. I'm sharing it because I think other developers - especially solo devs and indie hackers - hit the same problems I did. If you're running AI coding agents one at a time and don't have your tools setup in an optimal way, you're leaving a lot on the table.
macOS and Linux. Windows is on the roadmap.
Happy to answer any questions about the tool, the technical decisions, or why I chose to make it free.
I’ve tried using Claude and Codex together, but they often lose context when switching between them.
Curious how Parallel Code handles this. Do agents share context or is everything isolated per branch?
The git worktree isolation is the real killer feature here. I've been juggling tmux panes with Claude Code and it gets messy fast once you go past 3-4 parallel tasks. Having built-in diff review on top of that is a nice touch - that's usually where I lose the most time. Gonna try this out today.
Nice work shipping this. Running agents in isolated worktrees is the right call. The next 48 hours are when you find out what breaks under real load. Different OS environments, edge cases in the worktree management, sync issues. There's a free community stress-testing page where you post your URL and people try to find bugs. Findings come with screenshots. Saved me a few fires.
I've been doing this manually with tmux panes and git worktrees, so seeing it packaged as a proper tool with diff review is nice. The "zero switching cost" angle is key -- keeping agents as raw terminal CLIs instead of wrapping them in some custom UI means you're not betting against the pace these tools evolve at. What does the mobile monitoring actually show you -- just status, or can you intervene mid-task?
Running Claude Code and Codex side-by-side has become my default for anything non-trivial - they catch different things and the diff between their outputs is often the most useful signal. The context handoff between models is where it gets tricky, especially when they diverge on architecture decisions.
The isolated git worktree approach for running 10 agents in parallel is genuinely clever — no conflicts because each agent works in its own branch is the kind of thing that sounds obvious in hindsight but takes real thought to execute well.
I'm building CVDebug, an ATS resume scanner with Robot View that shows candidates exactly what Workday, Taleo and 50+ ATS bots see when they parse their CV, with an ATS score 0–100. Very different space, but I love the MIT open-source angle — transparency in how tools work is something I try to bring to the resume/hiring side too. Great launch Johannes!
This looks great for solo devs. Any plans to support team setups where multiple people dispatch agents to the same repo?
40+ keyboard shortcuts?? ok you have my attention 😂 I'm building a productivity app myself and I'm obsessed with keyboard-first workflows — reaching for the mouse when you're deep in flow is painful. The Alt+Arrow navigation between panes sounds super smooth. Definitely trying this out, congrats on the launch! 🚀
The automatic git branching is a solid idea, but how do you handle potential merge conflicts when multiple AIs are working on the same codebase?
Why I built Parallel Code
I've been maintaining an open source productivity app (Super Productivity) for 9 years and working as a programmer for much longer than that. When AI coding agents came along, they changed how I work - but the other tools and habits I grow comfortable with over the years, weren't a good fit anymore.
One agent works on a feature. I wait, get distracted, start something else. Finally, I come back, review, start the next one. Wait again. I had a backlog of ideas and the tools to build them, but things got messy. I was doing a lot, but I felt like I'd lost control compared to how I used to code.
I tried the obvious fixes. Multiple terminal windows were my new go-to solution, but got chaotic fast - which terminal was working on what? Tmux helped organize it, but I was still cycling through panes, switching tools for reviewing diffs and history, losing track of context, and dealing with merge conflicts when agents touched the same files.
So I built Parallel Code to fix this for myself.
The core idea: every task gets its own git branch and worktree automatically. Agents work in complete isolation - they can't conflict. I kick off 5, 8, 10 at a time, watch them all in a tiled view, and review diffs as they finish.
One design decision I'm particularly proud of: it runs your real terminal CLIs. Claude Code, Codex CLI, Gemini CLI - whatever you use today works exactly the same inside Parallel Code. Same behavior, same config, nothing changes. I didn't want to build another wrapper or chat UI. If you don't like the app, go back to raw terminals. Zero switching cost.
Some things I added along the way also turned out very useful:
- Phone monitoring - Scan a QR code, watch all your agent terminals from your phone. Sure there are other ways to do this, but this is really comfortable.
- Keyboard-first design - 40+ shortcuts. When you're managing 10 agents, reaching for the mouse is too slow. Best part is to navigate panes with just Alt+Arrow-keys.
- Built-in diff review and merge - The merge step is where quality happens. So having a quick way to do a code review is very useuful.
Parallel Code is free, open source (MIT), and always will be. No subscriptions, no platform fees, no telemetry. Your API keys stay on your machine.
I built this because I needed it. I'm sharing it because I think other developers - especially solo devs and indie hackers - hit the same problems I did. If you're running AI coding agents one at a time and don't have your tools setup in an optimal way, you're leaving a lot on the table.
macOS and Linux. Windows is on the roadmap.
Happy to answer any questions about the tool, the technical decisions, or why I chose to make it free.