A kanban board on your Mac that works its own cards.

What it is

• A board on your own Mac. Your cards sit in four columns: To do, In progress, Needs review and Done. It opens in your browser.
• Each card gets its own worker. A worker is Claude Code, working in a separate copy of your code. Several can run at once without getting in each other’s way.
• You stay in charge. A worker’s result waits in Needs review. Nothing is done until you accept it.
• It stays on your Mac. Your board, your code and the workers’ transcripts never leave it. The Vibe Hero service only sends you updates and takes the feedback you choose to send.

What you need

• A Mac with Apple silicon (M1 or newer). Intel Macs are not supported yet.
• Homebrew, the Mac’s usual installer. Get it at brew.sh.
• Git and Apple’s command-line tools. Setup checks for both.
• Claude Code, signed in to your own Claude plan. An API key works too, if you would rather pay per use.

Docker, Xcode simulators and the Android tools are optional. Setup tells you what each one adds.

Install it

Open Terminal on your Mac and run these three commands, one after the other.

1. Add the Vibe Hero tap to Homebrew:

   brew tap vibe-hero-app/tap

2. Install Vibe Hero:

   brew install vibe-hero-app/tap/vibe-hero

3. Set it up:

   vibe-hero setup

Setup checks your Mac, starts the board and opens Getting started in your browser. That walks you through the rest. There is no invite code: you sign up with just your name.

After that, the board checks for a new version every few hours and updates itself with Homebrew.

Privacy and safety

Your code, cards and transcripts stay on your Mac, in your own folders and the board’s database. Nothing is uploaded to Carlos or to any Vibe Hero server except what is listed below. The one exception you already know: what a worker, topic or chat reads to do its job goes to Anthropic through Claude Code, as it does when you run Claude yourself.

Everything that leaves your Mac

• Claude, at Anthropic
  Every worker, topic and chat is Claude Code, signed in as you.
• Feedback you choose to send — only when you send it
  Only what you write, only when you press Send.
• The update check — installed boards only
  Once, when you first save your name, the board signs up with the Vibe Hero release service: it sends that name and is given a username and a token. Every few hours after that it asks whether there is a newer version, with the token. An update is installed with Homebrew. If a new version fails to start, a problem report goes to Carlos: your username and name, the version and the last lines of the board’s log.
• The Claude status page — only while Claude seems down
  status.claude.com is read while Claude seems to be down, to tell an outage from a fault.
• The idle usage probe — only while idle
  A tiny Claude request, at most every 15 minutes while nothing else is running, to read how much of your plan’s usage window is left.
• GitHub and network probes
  A name lookup of github.com and one.one.one.one, to tell “no network” from a real failure. For a project with a git remote: a check that pushing to it works, a push of each card’s branch as a backup before it merges, and of the merged work once you accept it.
• Security scanners’ downloads — only when security review is on
  Semgrep’s rules from semgrep.dev, osv-scanner’s vulnerability database, and Safe Chain’s list of malicious packages.
• ssh to a deploy host — only when a project names one
  To deploy, and to check what production is running.

Workers

Each card’s worker is Claude Code in auto mode, working inside your repo in a folder of its own: it reads, edits and runs commands without asking about each one, while Claude Code’s safety check and the board’s guard decide what it may do.

Don’t ask for permission (bypass permissions) is off by default, on every card, goal and project. Switching it on turns Claude Code’s safety check off for that work, and the commands you marked “ask me first” run without asking. The never list and the board’s own token stay refused.

Your Claude plan

The board runs on your own Claude plan, through Anthropic’s own Claude Code. If you would rather keep its use off your plan, sign Claude Code in with an API key instead and pay per use:

claude auth login --console

macOS folder prompts

• macOS may ask whether node can read a folder. Say no: that is the board itself, and it never needs your private folders.
• Keep your repos out of Documents, Desktop and Downloads. The board refuses projects there.
• Topics run as their own claude and ask for folders too. Allow only the folders a topic should read, and say no to the rest.

What the board does to your Mac

Things the board can do to your Mac by itself. Each is off until you turn it on.

• Quit Docker Desktop when no card is using it
  After ten minutes with no card running in a project with Docker files. A container you started yourself keeps it running.
• Archive old transcripts every day
  Packs the Claude transcripts of the board’s own finished runs into an archive and removes the originals. Your own Claude sessions are never touched.
• Clone Android emulators for workers
  Copies the emulator a project names, so each worker has its own. Without it, Android cards get no emulator.

Feedback

Your board has a Feedback button. Write what went wrong or what you would like, add a screenshot if it helps, and press Send.

It goes to Carlos, who makes Vibe Hero and reads every one. Your board shows where each item stands: delivered, looked at, closed with a reply, or shipped in a new version.

Only what you write is sent, and only when you press Send. The one thing sent without asking is a short problem report if an update fails to start.

The fine print

Using Vibe Hero means agreeing to the user agreement. It is short and in plain English. Vibe Hero is built with other people’s software, listed in the third-party notices.