Corgi

Corgi is an open-source terminal multiplexer built for running AI coding agents alongside regular terminals. It is powered by Ghostty and inspired by herdr, and supports macOS, Linux and WSL. Running corgi in a project directory opens a workspace where agents such as Claude Code, Codex, Pi, OpenCode and Oh My Pi can be started and monitored.

What It Is

Corgi works like tmux for agent-driven development: agents and processes keep running in the background, and you detach and reconnect by running corgi again. A sidebar shows each agent's state (working, needs you, finished, idle), so you can see which agents need attention. Corgi detects agents run in any pane and installs hooks for Claude Code and Codex, or an extension for Pi and Oh My Pi. OpenCode needs no setup.

Monitoring and review

For supported agents, Corgi shows activity summaries, subagents, context, token use and rate limits. It can rename, fork and resume agents after a crash or reboot, and supports multiple accounts for Claude Code, Codex and Pi. A token view lists per-agent usage, and a resource view shows live CPU, memory, disk I/O and GPU use per workspace. A changes view lists the files an agent touched, and a worktree view compares branches against local main with uncommitted change counts.

Terminal features and scripting

Corgi supports keyboard and mouse use with a prefix key (Ctrl+B by default), tab and pane management, and undoing tab closes. Image paste works over SSH with Ghostty 1.4, and the project says it keeps pace on slow links. From any shell, commands like corgi list, corgi tab new, corgi pane split, corgi pane send and corgi pane read control the running session, with --json output. corgi --skill prints instructions that teach agents to use these commands.

Open-source model

Corgi is released under the MIT License. The interface and installer support English, Simplified Chinese, Japanese, Korean and Russian.