A terminal multiplexer keeps many agents alive at once

A terminal multiplexer runs many terminals in one window and keeps them running after you detach. tmux and Zellij have done this for shells for years; the gap for agents is that a plain multiplexer does not know which pane is an agent or what state it is in, so you are left stitching together bells and hooks yourself with no shared view of the fleet.

An agent-aware multiplexer closes that gap. Each agent gets a real terminal pane, the sidebar rolls every agent up to blocked, working, done, or idle, and the whole session survives a closed laptop or a flaky SSH link so you can reattach from another box or a phone. It is a single binary that runs anywhere you can SSH, not an app tied to one machine.

The utility is parallelism without babysitting: run a herd of coding agents at once, walk away, and none of them dies while you are gone. The session layer is part of the harness, the environment that keeps agents running (see: an agent harness is everything that constrains and shapes an agent), and the blocked, working, done view is observability at the session level (see: agent actions and planning must be observable). Because the multiplexer exposes a socket API, the agents themselves can drive it, spawning panes and reading each other’s output.