zerobrew claims up to 100x faster Homebrew installs using uv-style architecture

A new package manager called zerobrew has reached Hacker News's front page with an eye-catching pitch: install the same packages Homebrew does, but up to 100 times faster. According to the project's GitHub README, zerobrew brings a "uv-style architecture" to Homebrew packages on macOS and Linux, borrowing the approach that made Astral's uv installer a byword for speed in the Python ecosystem.

What the benchmarks claim

The project published a comparison of zerobrew 0.3.5, described as a development release, against Homebrew 7.0.8, measured on 8 October 2026. The test machine was a MacBook Pro with an M3 Pro chip and 18 GB of RAM running macOS 26.6.2 on a roughly 318 Mbit/s connection, with Homebrew starting from a completely empty state.

Across 100 packages, total cold install time — package and dependencies uninstalled, download cache empty — came to 775.6 seconds for Homebrew versus 117.4 seconds for zerobrew, a 6.6x speedup. Warm installs, where packages are removed but downloads remain cached, were far more dramatic: 637.5 seconds for Homebrew against 9.3 seconds for zerobrew, a 68x difference.

The median package was 5.3x faster cold and 56x faster warm. Individual cold results ranged from 1.6x (go) to 20.4x (ca-certificates), while warm results stretched from 17.9x (go) to 252x (ca-certificates).

Where the 100x figure comes from

The asterisk on the "up to 100x faster" tagline refers to a specific slice of the results: 24 of the 100 packages installed 100x faster or more in warm conditions. Node managed 121.7x warm, ffmpeg 107.4x and p11-kit 248.5x.

Cold installs tell a more modest story, and the project is candid about why: they are bound by the network. On a roughly 69 Mbit/s connection, the overall cold speedup fell to 3.3x, while the warm figure held at 69x. Large downloads such as go and llvm spend nearly all of their cold install time transferring data, so both tools converge there — llvm was only about 2x faster cold and go 1.6x.

Commands and installation

zerobrew installs through a standalone script served from zerobrew.rs, which updates the shell configuration, or through a Homebrew tap (zerobrewhq/zerobrew/zerobrew). An earlier tap under the lucasgelfond account is no longer updated, and existing users are directed to migrate.

The command surface will feel familiar to Homebrew users: zb install handles one or many packages, zb bundle and zb bundle dump work with Brewfiles, and zb outdated, zb upgrade, zb uninstall, zb reset and zb gc cover lifecycle management and garbage collection of unused store entries. A zbx command runs a package without linking it. Brewfile support means an existing Homebrew setup can, in principle, be exported and replayed without modification.

The benchmarks are also scripted rather than hand-assembled: running just bench --full results/ resets zerobrew, uninstalls everything it placed into Homebrew and writes a results table, while a smaller parity check runs in CI.

Why it matters

Homebrew's install speed is a recurring tax on developer machines and CI pipelines, and uv demonstrated that package management gets dramatically faster when the architecture — caching, linking and install flow — is rethought rather than incrementally tuned. zerobrew is an attempt to port that lesson to the macOS and Linux command-line package ecosystem, where Homebrew has faced little serious competition.

The caveats deserve equal weight. The results are self-reported, run on a single Apple Silicon machine with a fast connection, the software is a 0.3.x development release, and the headline 100x applies only to warm installs of a subset of packages. But warm scenarios — reinstalling after a failure, restoring a machine or provisioning CI — are exactly where slow package installs hurt most, and a 68x overall improvement there would be significant if it holds up under independent testing.