Agent Multiplexer

Watch several AI agents work at the same time, side by side, like a control room for the coding agents on your machines.

Each pane is a live session with one agent — Claude Code, Codex, Pi, OpenCode, and anything else connected to your computers. Plan in one pane, let another run the test suite, and keep an eye on a third. When an agent stops to ask you something, it stops in its own pane instead of blocking everything else you were doing.

Where to take it next

• Match the split to the job. Two panes for a pair-programming setup, four for a small team of agents, one big pane plus a strip of logs. The layout is a few lines in a settings file, so you can rearrange it without rebuilding anything.
• Give each pane a role. Name panes Build, Review, Explore, then tell each agent its role in the first message. The label follows the panel title, so you always know which is which.
• Use it as a dashboard for long jobs. Point a pane at an agent that runs a nightly task and leave it open: this is a calm place to check in on work that takes hours.
• Make it a teaching setup. One pane drives, another explains what it is doing. Students watch the reasoning and the result at the same time.

What to reuse in your own board

• A layout that is yours to change. Panes are grid cells described in one small settings file — a label, a position, and the agent it starts with. Any board that needs a resizable arrangement can copy this instead of inventing one.
• Embedded agent terminals. Each pane anchors a real terminal session to a region of the screen, and the session survives reloads because it belongs to that pane and that agent. This is the part to reuse anywhere you want an agent working inside a piece of your UI.
• Sessions that remember themselves. Disconnecting a pane keeps the conversation; reconnecting resumes it, even after switching that pane to another agent and back.
• A layout that reloads itself. Edit the settings file, or let an agent edit it, and the board rearranges while it is open. Useful whenever an arrangement lives in a file.

How to use it

1. Pick your agents. You choose one agent per pane when you add the board; you can change any of them later.
2. Panes connect on their own. Each shows a live dot while its session is running.
3. Switch agents from a pane header. Only that pane changes; the others keep working, and switching back finds the earlier conversation again.
4. Disconnect with the plug icon when a pane is done. Its history is kept — hit Connect to pick it up later.
5. Edit the layout in the board's settings file (data/config.json): column and row sizes, the gap, and one entry per pane with its label, grid position and starting agent. Changes apply while the board is open.

Nothing to connect up front: the board lists the agents already available on your workspace computers. A pane can only do real work while its agent is online.

Good to know

• A session belongs to a pane plus agent pair, so using the same agent in two panes gives you two independent conversations.
• The agent you pick in a pane header wins over the default in the settings file, and is remembered across reloads.
• This board is for agents that run on your own machines — the ones you can type into. The built-in assistant is not listed, because it has no terminal.