Run multiple AI coding agents without them fighting over files.
Want to run Claude in multiple terminals? They'll overwrite each other's changes.
Git worktrees could help, but setting them up is painful:
- Create the worktree manually
- Copy over
.envfiles - Don't forget
.env.local,.dev.vars... - Install dependencies again
- Copy
.claude/settings.local.json - 10 minutes later, you're finally ready
agentree -b new-featureOne command gives you:
- ✓ New branch (
agent/new-feature) - ✓ Isolated worktree in
../myrepo-worktrees/ - ✓ All env files copied automatically
- ✓ Dependencies installed
- ✓ Ready to code in seconds
brew tap AryaLabsHQ/tap
brew install AryaLabsHQ/tap/agentreeRun multiple AI agents in parallel:
# Terminal 1: Claude working on auth
agentree -b fix-auth
# Terminal 2: Another Claude on UI bugs
agentree -b ui-fixes
# Terminal 3: Cursor adding tests
agentree -b add-testsEach agent works in isolation. No conflicts. Pure productivity.
-
Interactive mode (recommended):
agentree -i
Guides you through branch creation with all options.
-
Quick mode:
agentree -b feature-name
Creates worktree with smart defaults.
-
Cleanup:
agentree rm agent/feature-name
Removes worktree when you're done.
Here's my actual workflow from last week:
# 9:00 AM - Start refactoring auth
agentree -b refactor-auth
# Let Claude work on this big task...
# 9:05 AM - Meanwhile, fix that urgent bug
agentree -b fix-login-bug
# Different Claude instance handles it
# 9:10 AM - Update docs while waiting
agentree -b update-api-docs
# 11:00 AM - Review all three PRs separately📚 Advanced Usage
# Create from specific base branch
agentree -b feature-x -f main
# Skip environment file copying
agentree -b feature-x -e=false
# Skip auto-setup (dependency installation)
agentree -b feature-x -s=false
# Push to remote after creation
agentree -b feature-x -p
# Create and open PR
agentree -b feature-x -r
# Custom destination
agentree -b feature-x -d ~/custom-dir
# Remove worktree and delete branch
agentree rm agent/feature-x -RCreate .agentreerc in your project:
# .agentreerc
POST_CREATE_SCRIPTS=(
"pnpm install"
"pnpm build"
"cp .env.example .env"
)Agentree automatically detects and runs the right setup:
- pnpm/npm/yarn: Installs dependencies + build
- cargo: Runs
cargo build - pip: Installs from requirements.txt
- go: Downloads modules
🛠️ Installation Options
# Download latest release
curl -L https://github.com/AryaLabsHQ/agentree/releases/latest/download/agentree-$(uname -s)-$(uname -m) -o agentree
chmod +x agentree
sudo mv agentree /usr/local/bin/git clone https://github.com/AryaLabsHQ/agentree.git
cd agentree
make build
make installBash
echo 'source <(agentree completion bash)' >> ~/.bashrcZsh
echo 'source <(agentree completion zsh)' >> ~/.zshrcFish
agentree completion fish > ~/.config/fish/completions/agentree.fish🤔 FAQ
Q: What's the difference between this and regular git worktrees?
A: Agentree handles all the setup that git doesn't:
- Copies your environment files
- Installs dependencies
- Copies AI tool configurations
- Does it all in one command
Q: Which AI tools does this work with?
A: Any tool that edits code! Tested with:
- Claude Code (what I built it for)
- Cursor
- GitHub Copilot
- Cody
- Continue
- Any future AI coding tool
Q: Can I use custom branch prefixes?
A: Not yet, but it's on the roadmap. Currently uses agent/ prefix.
Q: Does it work with monorepos?
A: Yes! Run agentree from any subdirectory.
I switched from Cursor to Claude Code to save money, but Claude is slow. Like, really slow. 10-minute waits for complex refactors.
So I started running multiple instances. But they kept overwriting each other's work. Git worktrees seemed perfect, but the manual setup was killing my productivity.
Agentree was born from frustration. Now I run 4-5 Claude instances in parallel, each on their own branch, with zero conflicts.
Found a bug? Have an idea? PRs welcome!
See something that could be better? Open an issue.
MIT
Built with ❤️ and frustration by @aryasaatvik