Stop Waiting on Claude: Run Six AI Coding Sessions at Once with Maestro

You know the rhythm by now. You fire off a task to Claude Code, watch it work, and then... wait. Maybe you tab over to check email. Maybe you start reading the next feature spec. But you're not actually shipping anything, because your AI assistant is a strictly one-at-a-time operation. Maestro is a desktop app that asks a simple question: what if you just ran several of them in parallel instead?

What It Does

Maestro is a cross-platform desktop application that lets you run between one and six Claude Code sessions (or other AI CLI tools) simultaneously. The core trick is isolation: each session gets its own git worktree, its own branch, its own terminal with a full shell environment, and even its own port allocation if you're doing web development. That means six agents can chew through six different tasks without stepping on each other's changes or creating merge headaches.

Under the hood, it's a Rust and TypeScript application with native builds for macOS 13+, Windows 10+, and Linux. The interface is a dynamic terminal grid that adapts from a single pane up to a 2x3 layout depending on how many sessions you've got going. Within each session you also get iTerm2-style split panes (Cmd+D for vertical, Cmd+Shift+D for horizontal), so you can keep an eye on multiple outputs inside a single worktree. You can pick per-session modes too: Claude Code, Gemini CLI, OpenAI Codex, or a plain terminal.

Why It's Cool

The worktree-per-session model is the right call. Rather than trying to multiplex one working directory (which invites exactly the kind of file collision you'd expect), Maestro creates worktrees automatically under ~/.claude-maestro/worktrees/ and prunes them when you close a session. Each one is on its own branch. You get real code isolation, not the illusion of it, and a "Worktree" badge in the terminal header so you always know where you are.

The built-in MCP server closes the feedback loop. This is the detail that elevates it beyond "a grid of terminals." Sessions report their own state (idle, working, needs input, finished, error) through a maestro_status tool, and that status shows up as real-time indicators in the grid. Instead of babysitting six windows, you can glance at the layout and see which agent is stuck waiting on you.

The visual git graph keeps you oriented. Once you've got half a dozen branches in flight, it's easy to lose track of what's happening where. Maestro includes a GitKraken-style commit visualization with colored rails, a branch relationship view, and a commit detail panel with diffs—so you can see exactly which session is touching which branch.

Quick actions cut the ceremony. Each session can have custom action buttons, including built-in ones like "Run App" and "Commit & Push," plus your own custom prompts. It's a small thing, but it saves you from retyping the same instructions across sessions.

There's also light and dark theme support and terminal font customization, which matters more than it sounds when you're staring at six terminals at once.

How to Try It

Since Maestro is a desktop application, you'll want to grab a build for your platform. Head to the repository first:

https://github.com/its-maestro-baby/maestro

From there:

1. Check the releases or installation section for your platform (macOS 13+, Windows 10+, or Linux).
2. Launch the app and create your first session—pick your AI CLI mode (Claude Code, Gemini CLI, OpenAI Codex, or plain terminal).
3. Let Maestro spin up the worktree, then start assigning tasks. Add more sessions as needed, up to six.
4. Use the status indicators to see which sessions need your input, and the git graph to keep tabs on branches.

Full documentation lives at the project's docs site, and there's a Discord server and an X account if you want to ask questions or follow along. If you're building from source, you'll need Rust 1.78+ and TypeScript 5.0+.

Final Thoughts

Maestro is aimed squarely at developers who've already integrated AI assistants into their workflow and have hit the serial-execution ceiling. If you're running one Claude Code session at a time and it feels like enough, you probably don't need this yet. But if you've ever found yourself queuing up tasks because your assistant can only handle one at a time, the parallel worktree approach here is a genuinely practical answer. It's MIT-licensed and still young, so expect some rough edges—but the core idea is sound, and the execution details (status reporting, worktree pruning, the git graph) suggest the authors have actually used what they built.

Follow @githubprojects for more developer tools and open source projects.