Self-organizing AI agent orchestration on Claude Code. Designed to force-multiply solo agentic engineers.
Give it a goal. It figures out how to organize agents to achieve it.
curl -fsSL https://raw.githubusercontent.com/dmotles/sprawl/main/install.sh | shOr with Go:
go install github.com/dmotles/sprawl@latestBuild from source:
git clone https://github.com/dmotles/sprawl.git
cd sprawl
make buildcd your-repo
sprawl enter
sprawl config set validate "make test"
sprawl config set worktree.setup 'npm install && cp $SPRAWL_ROOT/.env .env'Use sprawl config show to view current settings, or sprawl config --help for all options.
This launches weave (the root agent) inside the sprawl TUI. Give it a goal and it self-organizes from there — decomposing work, spawning agents, and managing everything autonomously.
- Navigate agents: Use the TUI tree panel to watch agents work; child agents still run in tmux windows you can attach to.
- Context handoff: When weave's context window fills up, ask weave to call the
handoffMCP tool. The TUI auto-restarts the session with memories of what happened. - Shut down: Ask weave to make sure no agents are running, then exit the TUI.
All of sprawl's state lives in .sprawl/ (gitignored), except for .sprawl/config.yaml which is tracked — it holds project-level settings like the validation command. Back up this directory to migrate between hosts.
- Claude Code
- tmux
- Go 1.25+
- Git
- DESCRIPTION.md — architecture and design
- CONTRIBUTING.md — how to contribute
- LICENSE
