Corgi is a terminal multiplexer for coding agents: it keeps sessions alive, labels each agent working, waiting, finished or idle, and replays the files it changed. It installs with one bash line and supports macOS, Linux and WSL (source).

Key facts

Fact Detail
Question answered Corgi: A Terminal Multiplexer Built to Watch Agent Tabs
Short answer Corgi is a terminal multiplexer for humans and agents: it shows which agent is working, which needs you, and which finished, keeps sessions alive in the background like tmux, and replays every file an agent to…
Reading time ~3 min
Last updated 2026-10-24

The install is a single shell line:

curl -fsSL https://corgi.computer/install.sh | bash

The supported platforms are macOS, Linux and WSL. That last one matters on Windows: the installer is bash, so the native path is a WSL 2 shell rather than PowerShell, and the agents you launch there work against Linux paths.

What the status line actually tracks

The useful part is not that it draws panes. It is that the pane knows what the process inside it is doing:

  • Every agent gets a state — working, needs you, finished, idle — plus unread and clear variants on the agent-aware view.
  • F1 is the catch-up key: one screen for what changed since you last looked.
  • Ctrl-T jumps into the selected agent’s work, new or existing.
  • Hovering an agent shows recent activity, which the site frames as “top for your tokens” — a way to see where the spend went.
  • The replay view collects every file that agent touched into one scroll.

None of this requires the agent to cooperate through a plugin API. Corgi sits at the terminal layer, so the same view works for Claude Code, Codex, Pi and OpenCode, and the site says it uses your existing accounts rather than asking you to re-authenticate.

Persistence is the tmux half

Corgi borrows tmux’s core promise: close the terminal and the agents keep running. The site shows a terminal, a dev server and a group of agents all alive in the background, then reattached later.

Two smaller details are worth calling out because they show the target user is someone living in a terminal all day:

  • Ctrl-Shift-T undoes a tab close, which covers the case where you kill the wrong agent.
  • Cmd-V pastes an image even over SSH, with the caveat that it requires Ghostty Nightly.

There is also a corgi --skill entry point, described as a way to teach your agents how to drive Corgi. Combined with the workspace and worktree organization, the design assumes agents will be started, inspected and moved between branches many times a day.

What it does not solve

Corgi is a visibility and persistence layer, not a permission layer. It shows you that an agent is working and what files it touched after the fact; it does not decide whether the agent may write to a path or call a tool. If your concern is containment rather than awareness, this is the wrong layer — you still need sandboxing and an approval gate underneath.

The other limit is platform shape. Windows users get WSL, not a native Windows build, so agents that need to touch Windows-only tooling, registry state or native GUI applications are outside what this setup covers cleanly. If your work is Windows-native, run Corgi inside WSL for the Linux-side agents and keep the Windows-side automation in a separate, explicitly supervised process.

How to try it without disrupting your setup

  1. Install inside WSL and start one agent you already use, not all four.
  2. Keep your current terminal open for a day and compare: the test is whether the status line stops you from polling tabs.
  3. Turn on the replay view for a single repo and check whether the file list matches what you expected the agent to touch.
  4. Only then move your main workspaces over, and keep a plain shell outside Corgi as a fallback.

If you want a desktop agent that runs alongside this kind of terminal workflow on Windows, the current build is available at agent.aimoogh.com/download. OpenShell’s kernel-enforced runtime, how a coding agent disabled its own safety controls, and the Windows download.

FAQ

Is Corgi open source? The site does not state a license or link a repository, so treat it as a product with a public install script rather than assuming you can fork it.

Can two agents share one workspace? Corgi organizes workspaces and worktrees, but nothing on the site claims file-level locking. For parallel agents, separate worktrees remain the safer default.

Related reading: