Corgi: A Terminal Multiplexer Built for AI Coding Agents (Show HN)

Corgi is a terminal multiplexer whose defining feature is that it treats AI coding agents as first-class citizens of the terminal rather than as just another background process. The project was posted to Hacker News as a Show HN and reached 38 points with 5 comments, and its landing page at corgi.computer describes an agent-aware workflow with status indicators, per-agent token accounting, and persistent background sessions.

What Corgi is, in one paragraph

Corgi is a terminal multiplexer for humans and AI agents, distributed as a shell installer at https://corgi.computer/install.sh. It positions itself between a classic multiplexer such as tmux and an agent orchestration tool: it keeps long-running agents and dev servers alive in the background, but it also tracks what each agent is doing, how many tokens it has spent, and which files it has touched. The tagline on the project page is "Terminal multiplexer for humans (and agents)."

The installer is a single command:

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

Agent awareness: status states and the F1 catch-up key

Corgi's most distinctive claim is that it displays the live state of every running agent. The project page lists four statuses:

• working, marked with a play-style indicator
• needs you, marked with a filled circle
• unread, marked with a filled circle
• clear, marked with an open circle

The page states that pressing F1 catches you up on agent activity. This is the core ergonomic argument of the tool: when several agents run at once, the bottleneck is not running them but knowing which one is blocked and waiting for input.

Which agents Corgi supports

The Corgi landing page lists four agent integrations by name, each with a logo and outbound link:

• Claude Code (claude.com/claude-code)
• Codex (openai.com/codex)
• Pi (pi.dev)
• OpenCode (opencode.ai)

The page also promises that you can "use your existing accounts. All of them." That framing implies Corgi wraps or drives the agent CLIs you already have authenticated rather than requiring separate credentials, though the landing page does not describe the mechanism.

Token accounting and activity tracking

Corgi advertises a token-usage view described as "top for your tokens," answering the question "Where did the tokens go?" It also offers activity tracking: hovering over an agent shows what that agent did. Both features target the same problem — that agent spend and agent actions are usually invisible once a session scrolls past the viewport.

Sessions keep running in the background

Corgi inherits the central promise of a multiplexer: processes survive detaching. The page states that, like tmux, agents and processes keep running in the background, and that typing corgi puts you back into the session. The page illustrates a session containing a terminal, an agent, and a dev server, with a status line reading "agents + dev server, running in the background."

File changes and git worktrees

Two features address reviewing agent output:

• Corgi shows every file an agent touched "in one scroll," which the page refers to as "see their changes."
• Corgi supports worktrees, described with the line "Fantastic trees and where to find them," implying it manages or surfaces git worktrees per workspace so parallel agents do not collide on the same checkout.

Neither feature is documented in technical detail on the landing page.

Workspace organization and resource usage

The page lists two workspace-level features: "clever corgi," which organizes your workspaces for you, and per-workspace live resource usage. Resource usage per workspace matters when several agents, language servers, and dev servers compete for CPU and memory on one machine.

Terminal ergonomics: images, links, and Ctrl-Shift-T

Corgi leans on modern terminal capabilities. According to the page:

• Command-V pastes an image, even over SSH. The page notes this requires Ghostty Nightly, linking to the tip release at github.com/ghostty-org/ghostty/releases/tag/tip.
• Links can be copied and images can be seen inline.
• New tabs and new agents are created with Ctrl-T, which the page describes as jumping into work, new or old.
• Closed tabs can be undone with Ctrl-Shift-T.
• The interface supports both keyboard and mouse.

Corgi is scriptable and self-documenting

The page states that Corgi can be driven from any terminal, making it hackable, and that running corgi --skill teaches your agents how to use it. That flag suggests a machine-readable or instructional output intended to be consumed by an agent so it can operate the multiplexer itself.

What the Hacker News discussion adds

The comment thread is short and contains no criticism of the architecture. Three comments are worth noting.

One commenter reported a month of real use and framed Corgi as a replacement for an existing tool:

"Been using this for about a month, for me it's been a great replacement for cmux with good simple design"

The same thread contains a packaging request and a competing open-source project. One commenter asked for a Homebrew formula and linked nocurlbash.com, a site arguing against piping installers directly into a shell:

"please add it to brew, I don't like to use bash scripts"

Another commenter, who says they built something similar, open-sourced it and emphasized the design goal behind this category of tool:

"Looks awesome! Will give it a try. I've built something similar, just open-sourced it: https://exawatt.ai I'm big on agent ergonomics. Cool to see what others are doing in the space."

The recurring theme in the discussion is "agent ergonomics" — the idea that the hard part of running multiple coding agents is human attention management, not model capability.

What the source does not say

This article is based on the Corgi landing page and a five-comment Hacker News thread. The source material does not disclose Corgi's version number, license, implementation language, pricing, or supported operating systems beyond the shell-based installer. No benchmarks, token-cost measurements, or performance numbers are provided. The claim that Corgi replaces cmux comes from a single user comment, not from the project itself. Readers evaluating Corgi should treat the feature list above as vendor-described rather than independently verified.