Rabbitty Download
A Mac terminal for AI coding agents (Beta)

A calm home for your AI agents.

A native Mac terminal for your AI coding agents. Run Claude Code, Codex, and more side by side, each on its own branch, and see which are done, stuck, or waiting on you.

For developers who run more than one AI agent at a time.

Download for macOS (Beta)
$ brew install --cask mauscoelho/tap/rabbitty-beta

Apple Silicon & Intel · macOS 13+ · Beta

One developer used it to go from 0 → 22 merged branches a day.

Rabbitty on macOS: a sidebar of AI coding agents showing Done, Idle, Working and Limit status next to a terminal.

Every agent gets a row in the sidebar with a live status like working, done, needs you, or rate-limited, next to a real terminal. One window for all your agents.

Why Rabbitty

One window for every AI agent you run

The agents do the work. Rabbitty keeps them organized and tells you when they need you.

Stop babysitting your agents

Start an agent task and switch away. Rabbitty fires a native macOS notification the moment the agent finishes or needs your input, so you stop polling terminal tabs.

Always know what each agent is doing

Rabbitty automatically parses terminal buffers to detect if an agent is working, done, needs you, or has hit a rate limit. No shell plugins required.

Run AI agents in parallel, safely

Launch as many agents as you need. Keep them grouped in isolated worktrees so they build and test on separate branches without touching each other's files.

Works with Claude Code, Codex, and more

Rabbitty runs your local CLIs unmodified. Whether you use Claude Code, Codex, Antigravity, or custom bash scripts, they run in a real terminal. No wrappers, no proxies.

Your code never leaves your Mac

Your code and terminal buffers never leave your machine. Rabbitty is a native app built on Rust, GPUI, and Alacritty, not an Electron shell. Everything stays local.

Diffs and commits, in place

Review your working tree without leaving the terminal. A keyboard-toggled panel (⌘⇧G) shows syntax-highlighted diffs, expands or collapses each file, and lets you commit inline.

Where we fit

Not a terminal. Not an orchestrator.

Rabbitty sits in between: it runs the real AI CLIs you already use, and adds the status, notifications, and safety a plain terminal doesn't.

Traditional Terminals

iTerm · Warp · Alacritty · Ghostty · cmux

Powerful and fast, but completely blind. They execute your AI CLI like a standard script, leaving you to stare at a screen waiting out rate limits, reviews, and approvals.

Blind execution
Rabbitty

The Command Center

Native agent command center

Observes the shell directly. Keeps your official CLIs as they are and runs them inside a command center that handles notifications, menu-bar status, and rate-limit detection. It also guards every worktree: terminals close before delete, and unmerged work asks to merge or discard.

Command center + Safety checks

Orchestrators

Conductor · Superset

Full apps that hide the CLI entirely or proxy prompts. Designed for running dozens of agents in parallel, replacing your terminal workflows and control.

Full tool replacement
Rabbitty compared with traditional terminals and AI agent orchestrators
Capability Terminals
iTerm · Warp · Alacritty · Ghostty · cmux
Rabbitty Orchestrators
Conductor · Superset
Runs official AI CLIs unmodified (Claude Code, Codex) Yes Yes Often wrapped or proxied
Agent status visible without switching tabs No Yes, in the sidebar and menu bar Yes
Parallel agents in git worktrees, one command Manual git worktree Yes, the agent starts right away Varies
Worktree lifecycle managed (status, merge, safe delete) No Yes Partially
Your shell and terminal workflow preserved Yes Yes, it is a real terminal No
Local-first, no extra account Yes Yes Not always

Worktrees

The secret to speed?
Build in parallel.

Running several agents at once is the fastest way to build. Rabbitty puts each one in its own git worktree, so they can all work side by side without ever touching the same files.

  1. 1

    Press ⌘N to start. Describe the task in a sentence. Your agent gets a fresh worktree on its own branch. Your main checkout never moves.

  2. 2

    Bring your local files. The first time, pick the ignored files it needs to run, like .env. Rabbitty saves the list to a .worktreeinclude and reuses it every time after.

  3. 3

    Let it work. The agent builds and tests on its own while you start the next one.

  4. 4

    Delete when it's done. Rabbitty merges the finished work into main, or asks what to do if anything is left unmerged.

⌘N New worktree
fix the env loader

branch fix-env-loader · claude comes along

Files to carry over .worktreeinclude
.env
.env.local
config.json

Delete fix-env-loader?

merged clean · 2 commits ahead

Merge to main & delete Discard Cancel

Quick Switcher

Run five agents.
Jump to the one that needs you.

Press ⌘P to see every agent at once, ranked by how much they need you. The one waiting on a decision sits right at the top. Type a few letters or a number, press Enter, and you land on it.

  • 1

    Ranked by attention. Agents that need a decision rise to the top. The ones still working sit below them, and idle ones fall to the bottom.

  • 2

    Your ask on every row. Each agent shows the task you gave it and its current state, so you recognize it without opening it.

  • 3

    One key, then go. Type a name or number to filter the list, then press Enter to jump straight to that agent.

Go to an agent by number or name…
1

fix-sidebar-hover

2m Needs attention

fix the sidebar hover flicker on inactive tabs

2

theme-two-step-picker

41m Done

split theme selection into a two-step picker

3

queue-retry-logic

12s Running

add retry logic to the queue consumer

4

brave-otter

2h Idle

explore migrating the glyph atlas to term-render

The numbers

Built with itself, measured in git.

Rabbitty is built inside Rabbitty, by AI agents. The numbers come from our own git history.

0 → 22

Merged branches per day, before and after parallel worktrees. Same agents, same developer. The difference is parallelism.

9 days

From first commit to a public macOS beta, built by AI agents inside the app itself.

4–5

Agents one developer runs at once, each in its own git worktree, all visible in the sidebar.

Give your AI a calm place to work.

Rabbitty is free and takes a minute to set up.