Orchestrator: Run Multiple Claude Code Agents Side-by-Side with Git Worktrees

Orchestrator: Run Multiple Claude Code Agents Side-by-Side with Git Worktrees

A new desktop app lets you run up to 10 Claude Code agents simultaneously on the same repo, using git worktrees to prevent merge conflicts.

6h ago·3 min read·3 views·via hn_claude_cli
Share:

What It Does

Orchestrator is a new desktop application built during a YC interview. It solves a specific problem: wanting multiple Claude Code agents to work on different parts of a codebase simultaneously without stepping on each other's changes.

The core innovation is its use of git worktrees. When you create a new agent in Orchestrator, it isn't just a separate chat window. It gets its own isolated copy of your repository on a dedicated branch. This means Agent A can be refactoring the API in src/api/ while Agent B is building a new UI component in src/components/, all without file-locking or accidental overwrites.

The interface is built like tmux, with split panes letting you watch several agents work side-by-side. You get real-time observability into each agent: every tool call, file edit, and token count is displayed. Sessions are saved to disk, so you can close the app and resume later with the full conversation history intact.

Setup

You'll need Node 18+ and the Claude Code CLI installed and authenticated on your system.

git clone https://github.com/MatchaOnMuffins/orchestrator.git
cd orchestrator
npm install
npm run dev:electron

The app will open, ask you to select a git repository, and you're ready to go. Click "+ New" to create an agent, give it an instruction, and watch it start working in its own pane.

When To Use It

This tool is for complex, parallel development tasks where a single agent isn't enough. Think of it as a multi-agent project manager.

Orchestrator running 4 agents simultaneously

Perfect for:

  • Large Refactors: Spin up one agent to update types, another to fix tests, and a third to update documentation—all concurrently.
  • Feature Development: Have separate agents build the backend endpoint, the frontend form, and the database migration for a single feature.
  • Bug Squashing: Assign different, unrelated bugs from your issue tracker to different agents and monitor their progress in one view.
  • Code Reviews: Use one agent to implement a change and another, in a separate pane, to critique the code as it's being written.

It provides deeper, stateful observability than tools like Conductor, which are more focused on single, ephemeral tasks. The key limitation is that you need to build it from source, as it's not yet packaged for easy distribution.

Built With Itself

In a fitting meta-moment, the developer notes that "orchestrator was built using orchestrator." It's a powerful demonstration of the tool's own capability for managing complex, multi-faceted development work.

AI Analysis

Claude Code users should start thinking about **parallelizing agent work**. Instead of giving Claude one monolithic task and waiting, break the problem into independent subtasks that can be executed simultaneously. **Change your workflow:** For your next feature, write a `CLAUDE.md` that outlines the discrete components (e.g., 'API layer,' 'UI component,' 'integration test'). Use Orchestrator to launch an agent for each component with a specific instruction. You become a project manager, overseeing progress and merging the finished worktrees back into your main branch. This can drastically cut calendar time on large projects. **A key tip:** The git worktree approach is the secret sauce that makes this safe. Before Orchestrator, trying to run multiple `claude code` sessions in the same directory was risky. Now, you can prototype this pattern manually by creating your own worktrees (`git worktree add ../feature-a-branch`) and running a separate `claude code` session in each. Orchestrator just automates and visualizes this process.
Original sourcegithub.com

Trending Now

More in Products & Launches

Browse more AI articles