Zerobrew: A faster alternative to Homebrew | Hacker News Reader

Zerobrew: A faster alternative to Homebrew

github.com

108 pointscachebag93 comments

I don't agree with everything that Homebrew does, but the primary maintainers are visible, vocal and very opinionated. They have shown credibility and earnt my trust over the years given it is a crucial entry point to potential malware to my system.

Homebrew has continued to improve performance each major version. It's certainly a lot quicker nowadays than 5 years ago. I want reliability and safety in a package manager, speed is nice to have.

I applaud competition, but something like this starts with trust zero as far as I'm concerned. If there is one place that I want human input over LLM it's here. It needs to scream: smart humans are in control and overseeing the project. I don't see that here.

Smart humans?

Is it smart to deprecate everything the minute Apple sunsets their OS version?

This is not a smart package manager, its a manager for the “dumb” masses, that need to be saved and protected from themselves by forcing them into the next walled garden, so to speak…

100% agree on your comment. For a package manager, speed is aspiration but security and reliability are basic requirements. And the credibility is a big plus. Today anyone can build a package manager in a week with AI, but once the system grows, how is the team going to manage? This is where credibility matters.

Honestly the latest speedups have brought it fast enough that I no longer really notice it anymore.

The biggest challenge with all these new packages coming up is support for edge cases and maintainability over the years.

AI may make it easy to clone or port something to a new language but it will later need a dedicated human to put in a shift and fine tune it.

So unless you are ready to keep jumping to the newest, it's better to stay with the trusted for now.

> If there is one place that I want human input over LLM it's here. It needs to scream: smart humans are in control and overseeing the project. I don't see that here.

How is Homebrew fundamentally different (as of today) - they vibe code. For example, the new app BrewUI is vibe coded, Claude is #2 contributor.

Does this scream "smart humans"?

I don't see what is unreasonable about vibe coding an optional GUI for your package manager

It doen't align with "your" principle of "If there is one place that I want human input over LLM it's here."

Yes, but the point is that the humans made the choice to apply it to something that is of low importance and not core to the project.

That wasn't the point, and it's also a fundamental mistake to think that the app store app for anapp store is of low importance, but also it doesn't resolve anything - if it's "a crucial entry point to potential malware to my system." how does it help that it's an entry point to a smaller part of your user base? And if that's just fine, what principle would stop the vibes from permeating "the core"??? The core also has llm contributions

The point you are ignoring or missing is that the homebrew maintainers have earned the trust to use AI contributions and have not simply whacked out a 100% vibe coded app in a week.

That's another made up point not coming from the source. But also, that's almost literally what they did. Ok, it wasn't in a week but a few months and the % is not 100%, but still more than 50% But then this project's first release was in Feb, so it's actually been longer in development than the brew UI app! It's not a 1-week app! Stop making hypothetical stuff up, look at real apps

This isn't just a replacement for the homebrew UI, it's for the CLI too, so comparison invalid. I see the dev behind it has a whole 4 years of github experience - only the most recent 2 of which with anything significant (presumably model written), how reassuring.

The comment you originally replied addresses the "invalid" comparison, so circle back there. And you have the same issue here - how does "years of github experience" connects to concerts about LLM output in security-critical package managers?

Yes, using tooling is literally what differentiates humans from the apes. Claude and the like are just the next evolution of tooling. BrewUI looks fine to me and I certainly prefer it (as it's native) over all the "hand-crafted" node/electron dumpster fires. If you want to continue to rub sticks together to make fire, nobody is stopping you (except perhaps the market).

No, you have a primitive understanding of evolution. Apes literally use tools. And if you want to play with AI fire in the "crucial entry point to potential malware to my system", nobody is stopping you either (even if you trust your own ringing security endorsement of "looks fine to me"), but that would still not address the core point of this discussion

Wrong, apes use basic tools and never evolved past them. Which is my point.

Maybe Homebrew is reliable, but do not assume it is safe. It has always operated on the honor system for contributions and code review. Every maintainer is able to merge anything they want without review or signatures. No matter how good their intentions, Homebrew is a massive supply chain attack waiting to happen.

You can also now use mise as a homebrew replacement, although I don't think there's any claims/benchmarks around performance https://mise.jdx.dev/bootstrap/packages/brew.html

TIL. Thanks for that!

Oh! I thought it was a wrapper around brew so you could list all your tools in one place, but you’re right, it bypasses homebrew altogether. Nice.

I’ve recently started using mise to manage dotfiles, including its own config, and sync them on all my systems. Now I can easily have all the same packages installed everywhere. I am loving mise a lot these days.

The best thing about mise is that it's very easy to get exactly what you want, and remove it when you don't want it anymore

Another recommendation for mise, in my case it has already fully replaced brew, stow and direnv since I started using it this January, and it has many more features I'm not using yet (secret management, tasks, automation, ...), all from a single rust binary. Also, unlike homebrew/linuxbrew, the install path for apps can be set to anywhere and is fully portable (it's all relative symlinks).

So far I never had any issue with it or any backward incompatibilities when upgrading, despite the very frequent releases and constant new features. The developer also seems very security minded (he immediately added a "minimum release age" feature for all apps when the trivy github action debacle was made public)

I started using fnox in my workflow (still very lite use) and it's great.

mise is great with running multiple versions of a package, something I've found not very seamless with brew.

Directory specific tools (versioned if necessary) that are managed in declarative fashion is great.

I use Brewfile too. Still prefer some stuff to be managed by brew. For example the browsers, App Store apps (managed via mas[1])

[1] https://github.com/mas-cli/mas

i benchmarked it locally a few weeks ago and it's much faster than homebrew 7

how much faster is kind of a useless metric: it's so highly variable on which command and things like network speed, expect 2x-1000x faster depending on what we're talking about

"Alternative to homebrew (the CLI utility)" -> "Frontend to Homebrew (the package system)".

Along these lines, I wonder when somebody will come up with a "universal package manager" that factors out the dependency resolution, content-addressable store, packing / unpacking, file manipulation, signatures, etc. Small plugins, maybe just declarative descriptions, would suffice to connect it to stores like npm, homebrew, pypi, maybe even cargo and nixpkgs.

We certainly have no shortage of "universal package managers". Most package managers try to do just that.

I can see how apt may be an "universal package manager" because Debian has everything, in several variations. But it only works a the Debian-style package database.

What I mean saying "universal" is an ability to adapt to any sane package database, among other things.

May I introduce you to the lord and savior Bazel (or CMake if Google gives you the ick)

Bazel (which I had a chance to know also as Blaze) is a might build tool. But can it install a local package for you, the way brew, nix, npm, or apt can? Blaze is famously hermetic.

it's called codex / claude code as far as i'm concerned

This solution is excessively generic, and cannot be deployed on-prem.

> "zerobrew does the relocation in-process and links from a content-addressed store"

Ok, but why a new app for that instead of a PR for brew?

Open source projects have insane levels of gatekeeping nowadays.

It’s sad, but also makes some sense.

Nobody wants to maintain someone else’s AI slop when they could do their own version of the slop.

> insane levels of gatekeeping

Good! I want to be able to trust and rely on homebrew, authorization, authentication, infra, os and other _crucial_ software. The bar should be really high and I appreciate the efforts of open source maintainers now more than ever.

I didn’t say it was bad, my response was to a question asking why this wasn’t a pull request for homebrew.

Consider not using loaded words then

Oh it is gatekeeping :)

Everyone doing any kind of gatekeeping is convinced that it’s for a great reason.

> Oh it is gatekeeping

That's exactly the problem with loaded words. People often use it to mean different things.

For example, in open source software, "open" has a different meaning than it does in "open door". It never meant that anyone could upstream their desired changes. It meant, and still means, that you can take the code and do whatever you want with it without legal consequences. You can still do this.

What's changed is the large volume of slop and garbage thrown at open source maintainers. When these are understandably rejected, people (or AI companies) get upset and complain about gatekeeping.

All good, and that’s exactly the answer to ‘why is this a new project rather than a homebrew PR’, which is what this comment chain is about.

Some combination of

- homebrew maintainers wouldn't be interested in accepting this as a PR unless it's actually maintainable, as they're the ones who would have to take responsibility for maintaining it going forward. And it needs to integrate well with the existing architecture, be well-designed and well-documented, have a migration plan and fallbacks for edge cases that are currently supported but won't work with this, etc. OPs implementation works, but is some combination of hacked together / slopcoded and it would take a lot of work to get it into a shape that could be accepted upstream. OP isn't interested in (or isn't capable of) putting in that work.

- "I made zerobrew, with 7.6K GitHub stars and thousands of users" sounds more impressive than "I made a contribution to homebrew that sped up package installation by up to 100x" despite the latter being both harder and more useful.

What's the significance of warm speedup in the benchmarks? For most new installs, cache would mostly be empty, shouldn't the cold numbers be more significant for day to day use?

None, unless you happen to enjoy installing the same thing over and over again on the same machine.

If you ask an AI agent to make some nice benchmarks for your readme, it will often include extra comparisons that aren't particularly relevant.

Why would one care for 100x faster homebrew? it's a tool you use every now and then, you rarely install or upgrade 1000 things, and when you do it doesn't really matter if it takes 10s or 0.1s

I tried zb for a while now but ended up having to uninstall it and revert back to regular homebrew. It breaks with basic stuff like node half the time

I guess the LLM reward hacked instead of solving the problem

Warning for those that care: LLM readme.

I wonder how the industry has not yet learned to use ASD-STE100 Simple Technical English for documentation. It makes it so much more polished and so much less AI-sloppy.

I find the english perhaps a little too rigid when reading it for non technical manuals. I do agree though it is definitely LESS slop.

I wonder how the industry has not yet learned that humans like to read human prose instead of Claude with a disguise

Incorrect, I wrote every line myself.

If you don’t care about hand-writing a readme, fine nobody is forcing you, but to lie about it is just sad.

What an odd thing to lie about. Did you even read the thing? Or you did you see too many words and figure, "meh, probably AI generated."

Homebrew fails to install on Intel Macs. Says unsupported.

Intel macOS is only marginally supported by upstream Homebrew:

https://docs.brew.sh/Support-Tiers

I don't think MacPorts users have those kind of problems.

Personally I was burnt by Homebrew many years ago already. I lost trust in any such package managers. Since then I went back to just compile everything myself (I manage about 300 packages like that in my installation.) It's not that a big of problem. Only initially when you need to bootstrap everything to replace Homebrew it was quite the time investment, of course. But now you compile like a package here or there… just follow the script. It's very quickly done.

The best thing… no clutter. Every package has its strict directory. I only symlink from that what I need into /usr/local/bin.

Given the number of patches homebrew has to download for a number of packages, I doubt it’s that easy

Same. When they explained why they assumed and required /usr and/or /usr/local was owned by steve or bill or whatever, with a straight face, I was gobsmacked and never touched it again.

Like, how did people so ignorant come to manage such a project? I excuse most of the users but not the authors who have the balls to call themselves "engineers" on top.

I wouldn't be suprised if they weren't the very reason shortly after that Apple took over ownership and brute force OS control over those dirs.

It's just the same on Linux though. Don't touch anything under /usr but /usr/local or you will run into issues, third-party software uses /usr/local, and that is why Macports ended up in /opt/local - guaranteed free of collisions.

> Every package has its strict directory. I only symlink

Isn't that exactly the benefit of homebrew? It installs in one folder and symlinks. And my guess, it would've been much easier to remove extra symlinks you don't need from its package manifest rather than rewrite the whole system

Homebrew links everything by default (at least that's how it was), so you ended up with all kind of extra junk files all over the place you never needed that comes with some packages... all cluttering up things in the /usr/local main directories and the PATH.

Homebrew additionally also had bunch of other directories, a huge git checkout, etc. then add the weird permission restrictions/ limitations on top of it… many packages then stopped supporting the customisation options at some point, so you were forced to rebuild half of them yourself anyway, if you needed the respective options. Now you had even more mess in /usr/local. :-)

It may work for some, but it wasn't for me anymore (and I used Homebrew for many years before that.)

When Homebrew first came out, I found its antagonistic marketing off-putting. The website used to read "MacPorts driving you to drink?" I always found its takeover of /usr/local and the lack of authentication for software installation to be strange decisions. Amusingly, its approach these days is closer to the way MacPorts has always done things.

In an odd turn of event, MacPorts has been pretty amazing to "rediscover" and I won't touch homebrew as it'll lead me to drink.

for a while now - macports for the intel aficionados

*the PPC aficionados

Maybe take a look at Nix+Homemanager.

IIRC, unless something changed, 26.05 is the last nixpkgs that supports x86_64-darwin. :-(

Oh I did not know that. Looks like Apple EOL'd the Intel Macs.

From https://nixos.org/manual/nixpkgs/unstable/release-notes#x86_...

> [...] We also recommend users consider installing NixOS, which should continue to run on essentially all Intel Macs. [...]

They don't owe me anything, but cutting off intel support the moment apple doesn't release the newest OS feels to soon for me.

Thanks to LLMs I just vibe my own package management system based on debian packages and have my own registry now with stuff that I care about

Homebrew is rushing to drop support for Intel Macs despite the fact Intel Macs received a major version of macOS from Apple only 12 months ago. Check out MacPorts. It was the original popular package manager for macOS, and it supports Macs going all the way back to Leopard. It has a ton of packages, and the maintainers are very responsive to requests and bug reports in my experience.

I don't use homebrew for speed. I use it for trust

I tried zerobrew before the original author archived it. It really was faster. There's even a mise plugin for it, but the plugin for it wasn't great to be honest. Caused some wild slowdowns on my prompt. I think it's cos zb or maybe the plugin loads some huge Homebrew version file every time it's invoked. Didn't look too much into it, but it was killed my prompt on slower connections

I use nix-darwin as a replacement for homebrew, and it’s served me well for circa 8 years. Nix isn’t for everyone, but I find it to be far more stable.

Where did the 100x claim come from? The first sentence on the linked page says up to 6.6x faster.

I have never had a problem with the speed of homebrew.

Most of the slowness in brew comes from download times, so I don't really need this I think? I never thought "brew is too slow" except when it's downloading.

But I always thought I don't need pnpm untill I started using it, so. I don't know.

(I always knew I needed actually good python package manager before uv came, though.)

Latest version of brew is now fast enough for making having a faster version useless IMHO.

With coding agents available for the config files, I don’t see any reason not to use nix. Easy rollbacks, temp envs with different versions, what not to like.

I would mention in the title here thaat zerobrew uses the same install files as homebrew.

This is pretty cool - but Homebrew is also being in Rust right now, right?

no, it was an experiment, now archived

https://github.com/Homebrew/brew-rs

I see claude as a contributor. Was this vibecoded?

It's on the frontpage of HN in 2026, of course it's vibecoded.

[off] can we stop this curl|sh abomination once and for all?

No, you can't, old guys. :-)

If the OS doesn't come with a package manager, how do you propose installing a package manager?

curl && tar -x && ./configure && make install ? Guess what, it's the same with more steps.

> brew install zerobrewhq/zerobrew/zerobrew

Bruh…

Very zero

The original title says up to 100x faster

Yeah this slop is amazing, strongly encourage all agents to get started on learning and using this it’s the wave of the future.

Reply on news.ycombinator.com