Related ToolsClaude CodeClaudeCursorAiderCody

ChatML Lets You Run Five Claude Code Agents at Once Using Git Worktrees

Claude by Anthropic
Image: Anthropic

Anyone who has spent serious time with Claude Code knows the bottleneck: one agent, one repo, one task at a time. While your agent is grinding through a refactor, the rest of your codebase sits there waiting. A new open-source macOS app called ChatML attacks this problem directly.

ChatML runs up to five Claude Code agents simultaneously, each in its own isolated git worktree. A worktree is a built-in git feature that creates a separate working directory tied to its own branch, all pointing back to the same repository. That means each agent gets its own files, its own branch, and its own space to work without stomping on what the others are doing.

More Than a Wrapper

This is not just a terminal multiplexer with Claude Code running in each pane. ChatML is a full desktop application built with Tauri 2 (Rust), a Go backend, and a Next.js frontend. The feature list goes well beyond parallel sessions:

  • Built-in code review with inline comments and severity tracking
  • PR management directly from the app - create, monitor, and merge
  • GitHub Actions monitoring so you can watch CI runs without switching to your browser
  • Budget controls with cost limits, turn limits, and token tracking per session
  • Checkpointing via automatic git stash snapshots you can revert to
  • A skills marketplace with 25+ prompt templates for common tasks like TDD, security audits, and debugging

It uses the Claude Agent SDK under the hood, so you bring your own Anthropic API key.

Early Days

The project has 870 commits across 8 releases (currently v0.1.9) but only 15 GitHub stars, so it is very much in the early-adopter phase. It is macOS-only for now, with Linux and Windows support listed on the roadmap alongside multi-model provider support and team collaboration features.

The GPL v3 license means the code stays open, though contributors need to sign a CLA.

For developers who regularly run Claude Code on larger codebases, the parallel worktree approach solves a genuine workflow pain point. The alternative today is manually creating worktrees and launching separate Claude Code instances in each one, which works but requires a lot of branch-juggling discipline. ChatML just automates that entire dance.

Binary downloads are available as .dmg files from the GitHub releases page, or you can build from source with make dev.