Show HN: Corgi, a helpful {agent, terminal} multiplexer

What it claims

Corgi is a terminal multiplexer meant to run several AI coding agents alongside ordinary shell processes. Its landing page lists two status vocabularies. The first is a per-agent work state: working, needs you, finished, idle. The second is an attention state: working, needs you, unread, clear. F1 jumps to whatever needs catching up on. The page says agents run with "your existing accounts. All of them." It does not name which agent CLIs or providers are supported.

How it works

The page describes features but gives no architecture. The stated behaviours are:

• Persistence. Sessions keep running in the background "like tmux", and running corgi reattaches.
• Token view. A "top for your tokens" view shows where tokens went.
• Activity. Hovering over an agent shows its recent actions.
• Changes. A single scrolling view lists every file an agent touched.
• Workspaces. Live resource usage is shown per workspace, and workspaces are organized automatically.
• Worktrees. Git worktrees are supported, but the page gives no detail on how.
• Modern terminal support. Links can be copied and images display inline. Over SSH, ⌘V pastes an image, and the page says Corgi "keeps pace" on slow links.
• Input. Both keyboard and mouse are supported.

How to use it

Install:

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

Keybindings named on the page:

• F1: catch up on agents that need attention.
• Ctrl-T: open a new tab agent, or jump to existing work.
• Ctrl-Shift-T: undo a tab close.

For scripting, the page says Corgi can be driven "from any terminal." Running corgi --skill teaches agents to control it. The page does not show the output format or a command API.

Limits

• Image paste, including over SSH, requires Ghostty Nightly. Other terminals are not mentioned.
• Installation pipes a remote script straight into bash. The page offers no package-manager or source install.
• The page states no licence, source repository, version, supported operating systems or supported agent tools.
• It gives no figures for overhead, latency or token-tracking accuracy.
• It does not explain how token usage is measured across different accounts.