Skip to content
ML2
Multi-Step Agent SetupsTool#Local / BYO model#Open source#CLI

Corgi: watch all your coding agents in one terminal

Added NewCopyable prompt below

Published Last verified Released

Our verdict

Worth trying if you run several coding agents at once and keep losing track of which one needs you. Corgi marks every agent as working, needs you or finished, F1 jumps to the next agent waiting on you, and agents keep running in the background after you detach, like tmux. It is weeks old and built by one developer: an early Hacker News tester reported repaint and flicker bugs, it installs with a curl | bash script (no Homebrew yet), and image paste over SSH needs Ghostty Nightly. Keep tmux or your current setup as a fallback for now.

Who it's for
Builders who run two or more terminal coding agents (Claude Code, Codex, Pi, OpenCode) in parallel on macOS (Apple Silicon) or Linux
Leverage
Less time checking idle terminals: one view of which agent is working, done or blocked on you, with background sessions you can reattach to
Effort
Low
Date added

What it does

Corgi (tals/corgi, MIT, written in TypeScript on Bun) is a terminal multiplexer built on Ghostty with agent awareness, and its README names herdr as its main inspiration. When you run claude, codex, pi or opencode in a pane, Corgi adds its hooks to that agent's settings (Pi gets an extension; Codex asks you to trust the hooks once in /hooks) and then shows each agent's status, activity summaries, context, token and rate-limit usage, and every file it touched. Sessions survive a detach, and agents resume after a crash or reboot. A launcher handles multiple accounts for Claude Code, Codex and Pi, and you can work with git worktrees and per-workspace resource usage. A corgi command drives the running session from any shell (list panes, open tabs, split panes, send text, read output, with --json), and corgi --skill teaches your agents to use it.

How builders use it

  • Install with the official script, run corgi in your project folder and pick a shortcut preset (the default prefix is Ctrl+B, then ? for the shortcut menu).
  • Click New in the agent list to start Claude Code, Codex, Pi or OpenCode; press F1 to jump to the next agent that is waiting on you.
  • Detach from the shortcut menu to leave agents and dev servers running, then run corgi again to reconnect.
  • Script it: corgi pane split current --right --command codex, then corgi pane send and corgi pane read to hand one agent's task or output to another.

What to expect

Several coding agents running in parallel, with one screen that tells you which one to look at next.

Copy this: Install + script a second agent

# Official source: github.com/tals/corgi (MIT). macOS (Apple Silicon) or Linux (amd64).
# Read the script before piping it to a shell.
curl -fsSL https://corgi.computer/install.sh | bash

# Start in your project, then click New to launch Claude Code, Codex, Pi or OpenCode
cd ~/dev/my-project && corgi

# From any shell: open a Codex pane next to the current one and give it a task
corgi list
corgi pane split current --right --command codex
corgi pane send 3 'run the tests' --enter
corgi pane read 3 --lines 50

# Uninstall
curl -fsSL https://corgi.computer/install.sh | bash -s uninstall

Related discoveries

See the newest picks →
  • ToolMulti-Step Agent Setups

    LangGraph

    Low-level framework for stateful, controllable agent workflows.

    Effort: High#Open source

  • ToolMulti-Step Agent Setups

    CrewAI

    Build collaborative agent crews and flows in Python.

    Effort: Medium#Open source

  • ToolMulti-Step Agent Setups

    OpenAI Agents SDK

    A lightweight framework for multi-agent workflows with handoffs and guardrails.

    Effort: Medium#Open source

Want more like this?

Practical discoveries, without the noise.

Get the newsletterMore in Multi-Step Agent Setups