curl -sSL https://raw.githubusercontent.com/lucasgelfond/zerobrew/main/install.sh | bash

After install, run the export command it prints, or restart your terminal.

Join the Discord for support / discussion.

A fast, modern package manager.

zerobrew applies uv's model to Mac packages. Packages live in a content-addressable store (by sha256), so reinstalls are instant. Downloads, extraction, and linking run in parallel with aggressive HTTP caching. It pulls from Homebrew's CDN, so you can swap brew for zb with your existing commands.

This leads to dramatic speedups, up to 5x cold and 20x warm. Full benchmarks here.

zb install jq                   # install jq
zb install wget git             # install multiple
zb install --file Brewfile      # install from a manifest
zb bundle                       # shorthand for Brewfile in current dir
zb uninstall jq                 # uninstall
zb reset                        # uninstall everything
zb gc                           # garbage collect unused store entries
zbx jq --version                # run without linking

Create a plain text manifest (compatible with Homebrew's Brewfile) listing one formula per line:

# Brewfile
jq
wget
git

Blank lines and comments (lines starting with #) are ignored. Install everything in the manifest with zb install --file Brewfile or use zb bundle to read the default ./Brewfile.

• Content-addressable store: packages are stored by sha256 hash (at /opt/zerobrew/store/{sha256}/). Reinstalls are instant if the store entry exists.
• APFS clonefile: materializing from store uses copy-on-write (zero disk overhead).
• Parallel downloads: deduplicates in-flight requests, races across CDN connections.
• Streaming execution: downloads, extractions, and linking happen concurrently.

I spent a lot of time thinking through this architecture, testing, and debugging. I also used Claude Opus 4.5 to write much of the code here. I am a big believer in language models for coding, especially when they are given a precise spec and work with human input! See some of the discussion about this on Reddit that convinced me it was worth adding to the README. A lot of people I respect, including the developers of uv are doing similar sorts of development, I don't think this is a particularly crazy practice in 2026.

/opt/zerobrew/      # Data directory (default: $ZEROBREW_ROOT)
├── store/          # sha256-addressable packages
├── db/             # sqlite database
├── cache/          # downloaded bottle blobs
├── locks/          # per-entry file locks
└── prefix/         # $ZEROBREW_PREFIX (default: $ZEROBREW_ROOT/prefix)
    ├── bin/        # symlinked executables
    ├── Cellar/     # materialized packages
    ├── lib/
    ├── include/
    ├── share/
    └── opt/        # symlinked package directories
~/.zerobrew/        # $ZEROBREW_DIR (source code, default: ~/.zerobrew)
~/.local/bin/zb     # $ZEROBREW_BIN (binary, default: ~/.local/bin)

All variables are respected by both the install script and zb CLI:

• ZEROBREW_ROOT
• ZEROBREW_PREFIX
• ZEROBREW_DIR
• ZEROBREW_BIN

cargo build --release
cargo install --path zb_cli

./benchmark.sh                                # 100-package benchmark
./benchmark.sh --format html -o results.html  # html report
./benchmark.sh --format json -o results.json  # json output
./benchmark.sh -c 20 --quick                  # quick test (22 packages)
./benchmark.sh -h                             # show help

Experimental. works for most core homebrew packages. Some formulas may need more work - please submit issues / PRs!

zerobrew is dual-licensed, usable under both Apache OR MIT, at your choice.