[TUTORIAL] How to install Claude Code on Haiku – because apparently waiting twenty years wasn't enough

[Andrea]

Haiku, 2024.

The fastest, most elegant, most incomprehensibly underrated operating system in the history of modern computing continues its glorious tradition of not having software.

No email client that actually works. No video calling app. No office suite. No Mastodon client, no maps, no ebook reader, nothing. We do however have a bug tracker with tickets opened in 2007 still waiting for a reply, and that gives us a certain historical continuity.

But hey, at least the software we don’t have was written by human beings. And that, apparently, is what matters.

Meanwhile, on this very forum, a remarkably refined philosophical school of thought has emerged: software generated by AI is unacceptable. Garbage. Stuff for lazy developers. We are craftsmen. We write code with our hands, with our sweat, with our suffering, as God intended.

Beautiful.

Too bad that in twenty years this approach has produced, roughly speaking, three ports of already existing software and a GTK theme.

So today I’m answering the question that many of you have asked me in private, yes, in private, because apparently asking for help installing AI on Haiku is still something to be ashamed of publicly.

Here’s how to install Claude Code on Haiku.

1. Install Node.js and npm via pkgman

pkgman install nodejs20
pkgman install npm

Verify the version (must be ≥ 18):

node --version
npm --version

2. Configure npm for global installs without touching system directories

mkdir -p ~/.npm-global
npm config set prefix ~/.npm-global

Add to your ~/.profile:

export PATH=~/.npm-global/bin:$PATH

Then reload:

source ~/.profile

3. Install Claude Code

npm install -g @anthropic-ai/claude-code

4. Verify the installation

claude --version

5. Authentication

claude

On first launch it will ask you to authenticate via browser. If you prefer to use an API key directly:

export ANTHROPIC_API_KEY="sk-ant-..."
claude

Note: the native installer method (curl -fsSL https://claude.ai/install.sh | bash) does not work on Haiku because it downloads a Linux binary. The only practical path is npm, which is what is described above. If you also want ripgrep for better performance: pkgman install ripgrep.

P.S. If your hand-crafted software from 2009 still needs a port, AI is available for that too. No judgment.

[waddlesplash]

Your post has a number of childish “shade-casting” statements. But ignoring those for the moment, and looking at the most glaring inaccuracy:

@Andrea

No email client that actually works. No video calling app. No office suite. No Mastodon client, no maps, no ebook reader, nothing.

No office suite? So what’s LibreOffice then? No email client? So what’s “Icedove” or “Trojita” then? No Mastodon client? So what’s “Tokodon” then? And we have software for maps, ebooks, and other things too.

Also, I moved your post to from the “Native” section to the “Ports” section, as none of the software your post is about is a native app.

[waddlesplash]

@Andrea

apparently asking for help installing AI on Haiku is still something to be ashamed of publicly.

Speaking for myself, at least, I would not mind at all if that was or remained the case, yes!

So far our rules against AI have been primarily about contributions to Haiku itself, with some discussion now happening about making some sort of rules at HaikuPorts. I would personally at least not be opposed to more restrictive rules at least about the “promotion” of AI.

Haiku has always been about “doing things right”, not “doing them fast”. If we just wanted “something that worked”, we’d have all given up and switched to Linux a long time ago. Haiku is a carefully designed and implemented system precisely because of the slow and careful work done by real, live human beings over the years. Practically all the things that people like about Haiku are or would be directly undermined by using “AI” to work on it (or for that matter, on third-party native software for it.)

So, yes, having software written by human beings is, in fact, what matters. Either “AI” is just a machine, in which case you’re asking a machine to do a human’s job, because programming is more an art than a science; or “AI” is sentient (which I don’t believe for even a second, but let’s suppose for the sake of argument that it is), in which case you’re advocating for software built by slaves. (Well, actually, you might be asking for software built by slaves anyway, knowing how sweatshop labor is often involved in the training of “AI” models… but I digress.)

Neither of those is a desirable outcome. I would rather see Haiku die than compromise on this point. And I don’t think I’m the only one.

[humdinger]

@waddlesplash

So, yes, having software written by human beings is, in fact, what matters.

Hear, hear.

It’s like bringing an AI that specializes in producing classical music to a philharmonic orchestra. “Now you can put down the instruments you’ve perfected together for most of your life, and listen to this. <press play on tape>”.

(For the record, I feel that I made a mistake supporting the development of EmailViews. I wasn’t aware at the start how much it relied on AI, and was frankly happy to see an email app that works how I’d imagine it. But after some soul searching, I couldn’t justify the damage that AI wrecks, and stopped.
What I don’t regret is the cooperation with ilfelice, always fun and open-minded.)

[Pap]

No, thank you.

This is not about the “tutorial”, really. It’s rather a LLM lover’s manifesto, full of inaccuracies and flat-out lies. It doesn’t take more than a second to realize the real purpose of this thread is to tell us how great LLMs are (and even worse, Claude, specifically). Even the title itself makes that clear. The “tutorial” is added just to make this “acceptable”.
It’s also crystal-clear that you think whoever isn’t a LLM-lover is “old fashioned” and “obsolete”. I could quote sentences about that but really, it’s not needed. The whole post is full of it, and it’s totally unacceptable.

I’d rather use my grey matter to write some useful code than correcting and “training” a dumb LLM so that, after a lot of effort and time, it finally gives a result that’s not complete garbage.

[nipos]

That “tutorial” is probably the biggest insult I’ve ever seen towards Haiku and its community
“No email client that actually works. No video calling app. No office suite. No Mastodon client, no maps, no ebook reader, nothing.” - That’s simply not true and you know that.
The email client included with Haiku works well for many people,and Beam works fine for many others.
Besides that,we even have a Thunderbird port based on the wayland compatibility layer.
It’s not native,but it exists and should work for a majority of email providers.
We have had LibreOffice for many many years,it was already there when I joined this nice community many years ago.
A native Maps application is available on HaikuArchives and many other things,like Mastodon clients,are ported from the KDE project which is possible due to our great Qt integration.
Ignoring the massive amount of work done by many volunteers to make Haiku a viable desktop computing experience and saying “all that will never work without AI” is just a massive insult to the work these people put in,a insult to the whole community and everyone who ever contributed to Haiku.
If I hadn’t already been in favor of banning AI before,I’d definitely be it now after reading this insulting post.
Thanks to @waddlesplash for making it clear that the rules are here to stay.
It seems at least the Haiku project isn’t going down the enshittification road,even if everyone else is.
Keep up the great human-made work!
I hope I’ll soon find more time again to contribute some human-made bugfixes,made with passion and love (and a lot of coffee) in my spare time.

[BiPolar]

@Andrea

[…] this topic is a tutorial and I’d rather keep it useful than turn it into a philosophy seminar.[…]

Totally agree. Then please remove the rage-bait, and honestly, pretty insulting, “intro” from it, and leave only the actual tutorial parts.

Edit: topic title too.

[Philipingram25]

I don’t mind using ai, but i prefer using human based but then again you had made the perfect beachmark I could add for ai comparsion for different OS so thank you but I did notice you threw shade at a lot of people

[humdinger]

Marking software as AI-assisted, with all the stigma this discussion implies, is not a quality label. It is a yellow star.

That’s quite an impropriate comparison! I’d rather liken it to a “fair trade” badge or for a product of sustainable fishery.

[Andrea]

@humdinger

That’s quite an impropriate comparison! I’d rather liken it to a “fair trade” badge or for a product of sustainable fishery.

@humdinger The yellow star was a deliberate provocation — and apparently it worked, since we’re talking about it.

My point is this: if we start labeling software based on the tools used to create it, where exactly do we stop? I’d like to know if a package was written on Windows, compiled with a specific version of GCC, or coded at 2am after three espressos. All of that influences the output too.

Jokes aside — a “fair trade” badge implies an ethical supply chain you can actually verify and trace. Who audits that? How do you prove a commit is “LLM-free”? The comparison breaks down quickly.

Judging software by its quality, security and license is already hard enough. Adding “how the developer felt about AI while writing it” to the list doesn’t make HaikuDepot better — it just makes contributing to it more uncomfortable.

[Andrea]

@Pap

No, thank you.

@Pap The tutorial works. I tested it on Haiku, the commands are correct, Claude Code runs. That’s the useful part.

The rest is my style, provocative, sure. “Manifesto”? If you like. But a manifesto that essentially says “here’s how to install a tool that might be useful to you” is a pretty peaceful one.

Nobody is forcing you to use LLMs. Nobody called you “antiquated”: I criticized an ideological stance, not people. If you recognized yourself in it, maybe that’s worth thinking about.

As for your grey matter: use it, it’s clearly what you do best. I’ll keep using every tool available.

[Andrea]

@nipos

That “tutorial” is probably the biggest insult I’ve ever seen towards Haiku and its community

@nipos I appreciate the passion, but it might be worth rereading the post a bit more carefully.

I explicitly thanked everyone who has done porting work over the years to make Haiku usable. That part apparently went unnoticed.

That said: LibreOffice, Thunderbird, the KDE clients you mentioned so proudly… do you know for certain that none of them contain a single line written with the help of an LLM? Because I don’t. And probably neither do you.

That was exactly the point of the post: AI is already everywhere, often without anyone knowing and without anyone raising an eyebrow. The idea of labeling software as “AI-assisted” on HaikuDepot is naive for precisely this reason.

And here’s a concrete example: Sestriere, the software I wrote to communicate with the MeshCore network, was written 100% with AI. I didn’t write a single line of code myself. I acted purely as a designer: I described what I wanted, evaluated the results, and asked for changes when something didn’t work. The software works. It’s useful. Does that count for something?

[Andrea]

@BiPolar

Totally agree. Then please remove the rage-bait, and honestly, pretty insulting, “intro” from it, and leave only the actual tutorial parts.

@BiPolar No

[Andrea]

@Philipingram25 Thanks! Discussion is essential, and sometimes a little spice does damage in the best possible sense: it brings out points of view that would otherwise stay hidden.

[nephele]

Jeez, you sure like digging a hole for yourself.

Feel free to read the forum FAQ : FAQ - Haiku Community

[Andrea]

@nephele

Jeez, you sure like digging a hole for yourself.

@nephele I read the FAQ carefully. My post criticizes an ideological stance, not specific people. No direct insults, no personal attacks. If there’s a specific rule you believe I violated, feel free to point it out: I’m happy to discuss it.

[nephele]

Strawman attacks to you are usefull “civilized” discussions I take it? There is no idealogical stance here. That would require an ideology. If you can’t cope with reasonable objections that is a you problem, not of the forum.

The forum before has had no Rules about discussins LLMs or “AI” tools. You just made that up. But considering the hostility of your post it may make sense to add such a rule in the future. Apparently people following the “AI Ideology” are just not civilized. (That is not serious by the way. There is no AI Ideology either, but you aren’t making any friends with your attacks)

Also, to my knowledge users have complained about you PMing them with AI stuff, not that they were not allowed to ask publically.

Also, the FAQ is deliberately vague, I don’t need to point you to section 1 paragraph whatever to see you clearly violated the very first sentence of the document, To keep this a civilized place for disccusions.

But hey, sure. Here is a specific rule you violated:

”Always Be Civil

Nothing sabotages a healthy conversation like rudeness:

• Be civil. Don’t post anything that a reasonable person would consider offensive, abusive, or hate speech.”

[Pap]

@Andrea

@BiPolar No

That “no” of yours simply proves my point: You just want to shout out how great LLMs are, as you did in the past and as you will surely do again and again, unless someone stops you. The “tutorial” is a perfect excuse to do so with yet another LLM’s greatness manifesto of yours nobody asked for.

@Andrea

But I’ll stop here, because this topic is a tutorial and I’d rather keep it useful than turn it into a philosophy seminar.

No, you didn’t stop there. Why? because, again, the point is to “educate” us how great LLMs are, by any means necessary. If you really wanted to “keep it useful” you would write your initial post in a completely different way. Without a long prologue telling us how great LLMs are and how Haiku is missing out — all that based in flat-out lies said in the most provocative way. You would just say something like “here is how you can use Claude in Haiku, use it if you need it”.
Oh, and if you really wanted to “stop there” you wouldn’t go on and on.

@Andrea

The rest is my style, provocative, sure. “Manifesto”? If you like. But a manifesto that essentially says “here’s how to install a tool that might be useful to you” is a pretty peaceful one.

I don’t think your so-called “style” fits with this forum. And that “pretty peaceful” part can only be interpreted as a (bad) joke. You clearly like arguing about LLMs — again, by any means necessary.

Feel free to keep going that way. Sooner or later, I guess someone would have to stop you — or not. Either way, I don’t really care. And I certainly don’t have any intention to feed this thread with more posts — but I’m sure you will.

[Andrea]

@nephele

Strawman attacks to you are usefull “civilized” discussions I take it?

@nephele Let me respond point by point.

Strawman attack: I criticized the idea of labeling software based on the tools used to create it. That’s an argued position, not a personal attack.

The PMs: what I shared privately is this: GitHub - atomozero/DenteBlu · GitHub — Haiku’s Bluetooth stack completed with the missing parts for pairing and communication with mobile phones. If anyone considers that spam, I’m happy to make those messages public.

The rule cited: “Don’t post anything that a reasonable person would consider offensive.” Would a reasonable observer find a tutorial with a provocative tone offensive, or would they find it offensive that in 2026 Haiku still can’t pair with a phone?