Concepts

You can use cadrei without reading this: run cadrei, then talk to your orchestrator. This guide is for when you want to know what happens underneath, shape your crew by hand, or fix something.

Your crew is your own folder of teams, members and settings. Older docs, and the folder layout (~/.cadrei/<name>, cadrei.conf), call it a cadrei.

Term	What it is
Crew	Your folder, ~/.cadrei/<name>, and a git repository. It holds your teams, members, playbook, projects list, settings and grants.
Your orchestrator	The Claude Code chat you talk to. cadrei opens it. It follows cadrei’s instructions (a Claude Code skill, a set of instructions Claude Code loads) and your playbook, and hands work to your team instead of doing it itself.
Team member	A long-lived Claude Code session with one role, started from members/<team>/<role>.md.
Team	A folder of members, started together or one at a time. A new crew has one: dev, with an engineer and a reviewer.
Project	A git repository your crew works on. The crew records its repository; each machine records where its folder is.

How cadrei is built

Cadrei works like Docker. The cadrei program is the engine: it starts and stops your team, keeps their permissions, and checks your setup. Your crew (~/.cadrei/<name>) is like a Docker volume: your members, playbook, projects list and settings all live there, in a git repo you can back up.

Each member file is like an image recipe, and a running member is like a container: start it, stop it, start it again, and it picks up its chat. Where your projects live on this machine is like a bind mount. Your crew keeps that list in a file that stays on this machine.

So the engine can come and go. Upgrade it, uninstall it, or delete ~/.cadrei/config and ~/.cadrei/framework, and your crew is still there. Run cadrei again and it plugs back in, with its chats right where you left them.