zerobrewhq/zerobrew: An as much as 100x* sooner Homebrew various · GitHub

curl -fsSL https://zerobrew.rs/install | bash

The installer updates your shell config. After it finishes, restart your terminal
or run the supply command it prints.

Or by way of Homebrew:

brew set up zerobrewhq/zerobrew/zerobrew

If you used the standalone installer, rerun it:

curl -fsSL https://zerobrew.rs/install | bash
zb --version

If you put in with Homebrew:

brew replace && brew improve zerobrew

If you put in from the outdated lucasgelfond/zerobrew faucet, it is not up to date. Switch to the brand new one:

brew uninstall zerobrew
brew untap lucasgelfond/zerobrew
brew set up zerobrewhq/zerobrew/zerobrew

zb set up jq                   # set up one package deal
zb set up wget git             # set up a number of
zb bundle                       # set up from Brewfile
zb bundle set up -f myfile     # set up from customized file
zb bundle dump                  # export put in packages to Brewfile
zb bundle dump -f out --force   # dump to customized file (overwrite)
zb uninstall jq                 # uninstall one package deal
zb outdated                     # record packages with newer variations
zb improve                      # improve all outdated packages
zb improve jq wget              # improve particular packages
zb reset                        # uninstall every thing
zb gc                           # rubbish acquire unused retailer entries
zbx jq --version                # run with out linking

Measured on 2026-10-08 with zerobrew 0.3.5 (improvement launch) and Homebrew 7.0.8 on macOS 26.6.2, MacBook Pro (M3 Pro, 18 GB RAM), ~318 Mbit/s obtain bandwidth. Homebrew began with nothing put in.

• Cold: the package deal and all of its dependencies uninstalled, empty obtain cache.
• Warm: the package deal and all of its dependencies uninstalled, downloads from the chilly run nonetheless cached.
• The median package deal was 5.3x sooner chilly and 56x sooner heat. Per package deal, chilly ranged from 1.6x (go) to 20x (ca-certificates) and heat from 18x (go) to 252x (ca-certificates). 24 of the 100 packages put in 100x sooner or extra heat, which is what the asterisk on the tagline refers to.
• Cold installs are certain by the hyperlink. The similar run on a ~69 Mbit/s connection (results/2026-10-07) got here out at 3.3x chilly and 69x heat. Big bottles like go and llvm spend practically all of their chilly time downloading, so each instruments land shut collectively there.

To reproduce, run simply bench --full outcomes/ on a machine with nothing put in in Homebrew. It resets zerobrew, uninstalls every thing it put in in Homebrew, and writes a desk to outcomes/benchmark.md.

A smaller model runs in CI: the parity workflow installs a hard and fast set of formulae with each instruments on the identical runner and compares the ensuing prefixes byte for byte, and the timing workflow occasions these formulae plus node and [email protected], chilly and heat, with each instruments and with the final zerobrew launch.

The nightly timing run fails if zerobrew is lower than 1.5x sooner than Homebrew chilly or 3x heat on any of them, or greater than 20% slower than the final launch general.

WRT Homebrew, I need it to be clear that each brew and zerobrew set up the identical bottles from Homebrew’s construct farm.

The distinction is what occurs after the obtain: Homebrew runs Ruby to guage the method, unpacks, rewrites paths with install_name_tool, and re-signs every binary separately, whereas zerobrew does the relocation in-process and hyperlinks from a content-addressed retailer, so it pays for the obtain as soon as and virtually nothing after.

I need it to be clear that we’re mainly nothing with out Homebrew’s bottle construct farm. Every package deal zerobrew installs is one they constructed, examined and revealed. We do not compile something, we do not preserve formulae, and none of those numbers exist with out that work. The pace comes from how the artifacts are put in, not from what’s in them.

This is an efficient learn, should you’re concerned with what zerobrew is and is not: Standing on the shoulders of Homebrew by Andrew Nesbitt. It makes the purpose higher than I can.

nanobrew:

nanobrew’s revealed numbers time a “heat set up” of a package deal that’s already put in, which is an early exit, not an set up, and examine in opposition to zerobrew 0.1.0.

Every quantity above is an actual set up from an uninstalled state, with the script and the total log on this repo, so you’ll be able to run it your self.

zerobrew is extra of a performance-optimized shopper for the Homebrew ecosystem. We depend on:

• Homebrew’s method definitions (homebrew-core)
• Homebrew’s pre-built bottles when out there
• Homebrew’s package deal metadata and roads

Our improvements give attention to:

• Content-addressable storage for deduplication
• APFS clonefiles for zero-overhead copying
• Source construct fallback utilizing Homebrew’s Ruby DSL

zerobrew is experimental. We suggest working it alongside Homebrew fairly than as a substitute, and do not
suggest purging homebrew and changing it with zerobrew until you’re completely positive in regards to the implications of
doing so.

• Status: Experimental, however fairly helpful. I (@cachebag) every day drive it myself.
• Feedback: If you hit incompatibilities, please open a problem or PR.
• License: Dual-licensed below Apache 2.0 OR MIT, at your alternative.