cookrew-app

agent
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • exec() — Shell command execution in cli/cookrew.mjs
  • process.env — Environment variable access in cli/cookrew.mjs
  • network request — Outbound network request in registry/assets/auth.js
  • crypto private key — Private key handling in registry/assets/device-seal.js
  • network request — Outbound network request in registry/assets/factors.js
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Open-source workspace where AI agents talk, drive browsers, and recruit help — terminals, notes, and live browsers wired together on one canvas. Cables are the permissions; every agent speaks one CLI.

README.md

Cookrew

Cookrew

A desktop app for running a crew of AI coding agents on your own machine. One view
zooms from your devices down to a single agent's terminal, your phone reaches the
same crew, and every agent gets a cookrew CLI to talk to the others.

cookrew.dev

Download

Get the latest release

Platform File
macOS, Apple Silicon Cookrew-<version>-arm64.dmg
Windows (preview) Cookrew-<version>-windows-preview-x64.exe

There is no Intel Mac or Linux release. Windows is a preview: agents end when the app
closes, and the cookrew CLI is not supported there.

Not notarized yet. Releases up to and including 0.5.1 are unsigned, so macOS
stops them on first launch. Until a release says it is notarized, drag Cookrew to
Applications and clear the quarantine flag before opening it:

xattr -dr com.apple.quarantine /Applications/Cookrew.app

An unsigned Mac build cannot install its own updates: it tells you when a new
release is out, in the ⋯ menu, and opens the download page. The 0.4.0 download
also leaves cookrew out of its terminals; 0.5.0 fixes that.

Requirements

  • macOS 11 or later on Apple Silicon.
  • Node.js on your PATH. The cookrew command runs on it.
  • At least one agent CLI: Claude Code, Codex, OpenCode or Pi. A plain shell works too.

Optional, and each one turns something on:

  • herdr or tmux: agents keep running when Cookrew closes.
    Install herdr with curl -fsSL https://herdr.dev/install.sh | sh. Without either,
    closing the app ends the agents.
  • Google Chrome, Chromium or Microsoft Edge: agents and the phone can drive
    browser cards.
  • Tailscale: the phone reaches this Mac away from home
    without the cookrew.dev relay.

Cookrew finds herdr and tmux on its own PATH, and macOS gives an app opened from
Finder or the Dock a minimal one that includes neither ~/.local/bin nor Homebrew.
To keep agents across restarts today, start it from a terminal:

/Applications/Cookrew.app/Contents/MacOS/Cookrew

The zoom view

The whole app is one zoom with named stops on the bar:

devices → workspaces → artifacts | agents

Click this Mac, a workspace or a card to go in. Other Macs on your account are listed
but open only on their own screen. Click the background, press Esc or use the back
button to come out. An agent opens from the crew into its card, then into a full view.
The last stop on the bar flips a workspace between its artifacts and its agents. On a
touch screen, pinch works too.

Press / to search everything on the board. Esc and / belong to the terminal or text
field while you are typing in one. The prompt box at the foot of the window
talks to the agent you have open or named with @agent. With nobody targeted, it
starts a new agent and names it from what you wrote. Hold ⌘ to dictate; speech is
recognized on your Mac.

Agents

An agent is a real terminal running a harness: Claude Code, Codex, OpenCode, Pi or a
plain shell. Its face shows whether it is working, done or waiting on you, and a chime
rings for those last two. The full view shows the same agent as a readable chat or as
the live terminal.

One agent in a workspace can be the orch, which leads the crew and may restructure it.

The Claude Code preset starts with --permission-mode bypassPermissions, and Cookrew
is not sandboxed: agents can do what your user account can do.

Artifacts

What agents produce and use, on the other side of the same zoom:

  • Browsers: live pages an agent can read and drive, with tabs.
  • Notes: markdown, stored as real .md files under ~/.cookrew/workspaces.
  • Screens and devices: a browser card can also show a Mac display or window
    (screen://main, window://App), an Android phone (adb://SERIAL), a virtual
    iPhone (vphone://VM) or a camera (cam://0). These cards need Chrome, Chromium or
    Edge installed, as above. An Android card needs adb, and a virtual iPhone needs a
    running VM. The Mac cards need Screen Recording permission, and each card says what
    it is waiting for and what to do next.
  • Remote desktop: the Remote desktop button on a machine opens its screen from the
    desktop or a phone. Cards are view only. Turning on Allow remote control of this Mac
    (in the ⋯ menu or on the device page, at the Mac only) lets a paired device take
    control of a card for 15 minutes, and lets agents click and type; both also need
    Accessibility. While anyone watches, the Mac shows who, with a Stop button, and stays
    awake. A locked Mac needs someone there: nothing here types into the lock screen.

Phone

cookrew mobile prints the address to open on the phone: no app to install. You get
the same zoom, terminals, the prompt box and dictation. Signed in with a Cookrew
account, the link goes through the cookrew.dev relay and works from anywhere, and your
profile shows it as a QR code. Without an account the phone connects directly, over
your Wi-Fi or Tailscale. cookrew mobile --rotate revokes the pairing token.

The cookrew CLI

Cookrew puts cookrew (or crew, for short) on the PATH of every terminal it runs.
It is the whole agent API, with no SDK to install.

cookrew list                               # who and what is wired to you
cookrew ask "Atlas" "why is CI red?"       # send a prompt, wait for the reply
cookrew ask "Atlas" "run the suite" --no-wait
cookrew status                             # thinking, waiting, replied or idle
cookrew check "Atlas"                      # read an agent's terminal
cookrew recruit --prompt "port the parser" # start a teammate, named from the prompt
cookrew note create "Release checklist"
cookrew browser create https://example.com "Docs"
cookrew browser snapshot "Docs"
cookrew browser create screen://main "My screen"
cookrew team save "Reviewers"              # snapshot the crew as a reusable team
cookrew login                              # sign this machine in, no window needed
cookrew whoami                             # the account and device it is signed in as

Run cookrew help for the rest. Asking or checking an agent, reading or editing a note
and driving a browser by name wires you to it. Verbs that restructure the crew, such as
connect, dismiss, team save and workspace switch, are for the orch.

cookrew login and cookrew logout are for you, not your agents: they run in a Shell
card, and the password is typed at a hidden prompt, never on the command line. Signing
in adds the machine to your account, so it is also how a machine with no window joins.

License

MIT

Development

npm install        # rebuilds node-pty for Electron, builds cr-listen
npm run dev        # the app, with hot reload
npm test
npm run typecheck
npm run dist:mac   # dmg and zip in release/

If node-pty fails with 'functional' file not found, your Command Line Tools are
missing their libc++ headers:

export CXX="c++ -isystem $(xcrun --show-sdk-path)/usr/include/c++/v1"
npx electron-rebuild -f -w node-pty

Releases come from .github/workflows/release.yml on a v* tag. It signs and notarizes
the Mac build when the secrets are set, and says so in the job summary when they are
not. To check a build the way macOS will, without launching it:

scripts/verify-mac-release.sh release/Cookrew-<version>-arm64.dmg

Reviews (0)

No results found