teamctl

Run real AI agent teams from one YAML. Each agent is a long-lived CLI process.

Long-running AI agents — Claude Code, Codex, or Gemini sessions — declared in a YAML file you write yourself, supervised on your machine. Each agent runs in its own tmux pane; they coordinate through a shared mailbox. If you've used docker-compose, the shape is similar — declarative, supervised, file under your version control. The manager pauses for you on anything that matters.

Interactive Setup (with Claude Code) - Recommended ⭐

claude plugin marketplace add https://github.com/Alireza29675/teamctl
claude plugin install teamctl@teamctl
# inside a Claude Code session:
/teamctl:init
# back in your shell:
teamctl ui

The plugin walks you through what kind of team you want, scaffolds a .team/ folder in your project, brings the agents up, and shows you everything in teamctl ui.

Manual setup 🔧

Install teamctl:

curl -fsSL https://teamctl.run/install | sh

Already running Claude Code? Run the installer through bash -c to be prompted to install (or silently update) the teamctl plugin in the same step:

bash -c "$(curl -fsSL https://teamctl.run/install)"

The pipe form (| sh) stays silent so CI workflows are unaffected.

Then bring up a team:

cd /to/your/project
teamctl init # initiates the .team/ folder in your project
teamctl bot setup # helps you connect 'managers' to Telegram bots
teamctl up # starts the team
teamctl ui # interactive team inspection

Learn more 🔍

• Documentation — guides, concepts, reference, ADRs
• Example teams — OSS maintainer, editorial room, indie studio, solo triage
• How teamctl compares — feature matrix vs neighboring tools

License

MIT