kumo

蜘蛛 — spider in Japanese.

Weave your AI agents into a single web

A featherweight terminal multiplexer built in Rust. Every pane is a real Ghostty-emulated terminal, every AI CLI you run is a thread in the web, and the sidebar always tells you what each agent is doing.

Get Started GitHub

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/marcrdgz/kumo/releases/latest/download/kumo-installer.sh | sh

Featherweight by design: kumo idles at ~0.7% CPU and ~6 MB of memory while you work. 🪶

Real terminal emulation

Each pane is a genuine VT/xterm emulator (vendored Ghostty) — shells, TUIs, and full-screen apps behave exactly like in a native terminal.

Split panes & sessions

Binary split tree, mouse-drag resizing, zoom, multiple independent sessions.

AI CLI panes

Spawn opencode, claude, codex… with one keystroke; kumo auto-detects agents in any pane.

Agent status at a glance

Green working, orange blocked, gray idle — detected from the live buffer, with audible alerts.

Orchestration

Agents drive their own web — wait, prompt, read, dev servers, and parallel workers, no polling.

Detach & re-attach

The daemon keeps your web alive in the background; come back from any terminal, exactly where you left it.

Configurable

Leader keys, bindings, shells, AI commands — all in one config.toml.