Multi-agent work quickly turns into too many terminals, copy-pasted context, unclear ownership, and no trustworthy view of who is doing what.
A project-local cockpit for running real coding agents side by side while preserving their native terminal behavior.
It adds coordination state, roles, task visibility, and MCP tools around the agents instead of replacing them with a fake chat UI.
Define roles, launch a session, let agents use shared MCP tools for messages/tasks/context, and watch their panes, git state, and project tasks in one TUI.
The output is a more controlled multi-agent work session with shared context and fewer human routing chores.