Why every terminal pane should be an AI agent

The terminal is where builds run, tests fail, and deploys happen. So why does the AI live in a different window?

Most AI dev tools are a chat box in a browser or an editor sidebar. They can suggest code, but they can’t see your terminal — the exit code that just came back, the stack trace, the half-finished git rebase. So you copy-paste context back and forth, and the model works blind.

One pane, two ways to work

Rysh takes a different stance: every pane is both a shell and an agent. In shell mode you get a normal, PTY-backed terminal. Double-press Escape and the same pane switches to prompt mode — now you’re talking to an AI agent that already has the pane’s context. The agent can read and edit files, run bash, drive git, build and test, using 40+ tools.

Because the agent lives in the PTY, it sees real output and iterates on it — no copy-paste, no lossy screenshots.

Multiplexing, kept

You don’t give up the terminal multiplexer to get this. Tabs, panes, splits, stacks, layouts and persistent detach/reattach sessions all work like you expect from tmux or zellij. The AI is additive: one agent per pane, coordinating across panes when you want it to.

The payoff

When the model shares your context and your tools, it stops being a suggestion engine and starts being a collaborator that can actually do the work — with you approving anything that touches the system.

Related: See it in a 60-second video · What is Rysh?