Use case
Put ten agents on it at once.
One window, up to sixteen AI coding agents — each on its own task, on its own model, in your own repo. You stop waiting on one assistant to finish before starting the next thing.
Why it’s for you
Parallel, not queued
One pane refactors while another writes tests and a third chases a bug. Each is a live session of Claude Code, Codex or whichever tool you pick, on your own machine, working in your real project at the same time.
Every plan you pay for, working
Queue the work and each task goes to whichever signed-in subscription has room, in its own git worktree. When one plan hits its usage limit, the task moves to the next, and you review a diff before anything is merged.
The right model per task
Each pane keeps its own model. Put the hard reasoning on a bigger model and the mechanical reformat on a cheap fast one — you're not paying premium rates for a rename.
Plain shells when you want them
Not every pane has to be an agent. Keep an ordinary terminal alongside them to run the tests and read the git status yourself.
Your voice, when typing is slower
On Windows, hold Alt and talk. Speech is transcribed on your device — the model runs locally, so what you say goes no further than your machine.