Projects, teams, and machines

Team — the billing and permissions unit. Your subscription, members, and plan live at the team level. Everyone on the team can be invited to individual projects.

Project — the unit of work. Each project has its own chat history, files, connected integrations, workflows, and one dedicated AI agent. Projects are isolated: a conversation in one project never sees another project's context.

Machine — every project runs on its own long-running machine. The agent, its filesystem, and its sandboxed browser all live there. You don't manage this directly — the machine starts when you chat and suspends when idle to save resources.

States you'll see — provisioning when a new project is being set up, running when the agent is active, suspended when it has gone idle (it wakes automatically on your next message), and stopped if something has gone wrong. You can see the current state in the project switcher and in the project's status panel.

Switching projects — use the project switcher in the app header. Each switch changes which agent you're talking to.

One agent per project, by design — Magister does not pool one agent across projects. If you run three brands, you run three agents, and each gets its own memory, connections, and credit usage.