I'd recommend the free (as in freedom) Git alternative to everyone: Codeberg. Super lightweight, without that vicious circle of knowing your code is being used to train AI. Free for hobby projects, and reasonably viable for professional ones.
Do they really prohibit all use? My reading of their policy is that they mainly block vibe coding. It doesn't seem like LLM auto-complete or using it to review your code would run afoul of it.
Hmm I have my Hugo website there, closed repo and build, in part, with LLMs. Never realized while setting thing up that both are not allowed tbh. I just wanted to not choose GH.
I wonder how much internet-argument bandwidth would be saved if we used different words for "freedom from" vs "freedom to", instead of putting both things on a single scale and then arguing over which is "more free"...
Insecure human slop on top of that. Projects not checked in-depth by AI is recklessness at this point, prove me wrong with an example.
I've found 2 serious vulnerabilities in Librewolf (can crash computer from HTML page, from an iframe) as an example this week, as per their rule, agent isn't allowed to report it, guess we will just have to leave those critical around or wait around that a human discover them while attackers already exploit them in the wild :/
Codeberg doesn't want the world to be secure, quite shady, unless they care to explain what's the downside of discovering critical vulnerabilities with AI, especially automatically? Thinking that humans will write more secure programs than machines is just wrong and shows an outdated and dangerous mindset.
Their ToS only forbids projects that 'mostly consist of code' created with genAI, it says nothing of code that is reviewed through it or projects where it only makes up a small portion of the whole package. The reasons for that are outlined and quite clear - their organization focuses on projects with a certain license and LLM code is in limbo regarding its copyright status.
Basically, unlike your stance, they are rejecting people who are outright saying that AI alone > humans. Which seems to be a popular opinion on HN. Given how angry you are about this, I don't think there's as much separating your opinion from the vibe coders' view as you think there is. I mean, look at this, we're only a few years into LLMs being viable code generators and we already have people accuse them of malice, recklessness and embracing 'human slop' because they dare to have a softball policy on not using this brand new tech absolutely everywhere and for everything. What will the average opinion be like in 2 years? Humans are just bad and can't be trusted to do anything manual at a computer, leave it to the agent?
So they will never have these frequent downtime issues because they will never be bombarded by agents building millions of personal tools with no license applied. Clever.
Free as in freedom has had a rather specific meaning in software for the past 40+ years, which is "software that matches the Free Software definition." Codeberg runs on Forgejo, whose authors grant you the rights to run, study, modify, and distribute it - unlike GitHub, which is proprietary. (Indeed, there are many Forgejo instances besides Codeberg, it's not just a theoretical benefit.)
It is a ridiculous misinterpretation to assume it means "somebody else will host anything I upload, free of charge", which is of course not the case - not on GitHub either, by the way, their ToS just happens to be more friendly to slop.
(For understandable reasons, too; while M$ is making a killing from selling LLM access & Free Software as training data, the main "benefit" Codeberg sees is getting DDoSed by LLM scrapers 24/7 and vibecoders using up disproportionate amounts of resources.)
People who migrate away from Github, what makes you think they're more reliable? All Git services are prone to massive surges in commit volume due to AI agents. At least if Github goes down, your boss is not going to blame you.
I'd rather stay in Github and maybe use another service purely as backup.
> All Git services are prone to massive surges in commit volume due to AI agents.
You are probably wrong on that.
> You must not share projects that mostly consist of code written by "generative AI"-tools (including services such as Claude, OpenAI Codex). Such projects having an unclear copyright status (see requirements § 2 (1) 1 and § 2 (1) 3) and furthermore have little safeguards to ensure that they do not include harmful code (c.f. § 2 (1) 5).
You missed the rest of the comment, which was to the effect that such AI agents are not allowed on Codeberg. Any abuse at scale will be easily detected and banned.
Are they going to ban your account based on an algorithm? Are they going to personally check your code? Are they going to scan your code base for AI generated code? How do they define “mostly”?
These seem like insane policies to have a critical business dependent on.
Not only that, their uptimes are still worse than GitHub.
> I'd rather stay in Github and maybe use another service purely as backup.
Swap those two and you're on the right track. I think the best way to use GH is as a mirror for a self-hosted git instance, with or without web frontend. It gives you the reach of GH with the, dare I say it, reliability of self-hosted. It also gives you the chance to keep things off GH which you don't want to be there. Keep the self-hosted instance private for most things, use GH links for public access. When GH is down again you can always give people (temporary) access to your self-hosted instance if the need arrives.
Why exactly do you think public code on Codeberg is not used to train AI? I would assume the opposite...it's unlikely to be Codeberg themselves doing it, but for sure your code is going to be used as training data somewhere.
TOS are not protection against scraping. Is there any reason to not assume by default that Codeberg gets used for training just like every other online collection of content?
"This endpoint is seeing a high influx of requests for this repository, to preserve the availability of Codeberg your request will not be processed. Sorry for the inconvenience and please try again later."
I've had issues recently with CI jobs in codeberg taking quite a bit of time for a runner to pick it up.
This is not a complaint, its a free offering and I appreciate alternatives, but just yesterday I moved one of my open source projects from codeberg to github after waiting 30+ minutes multiple times for CI to start.
Hosted services can't really be "free as in freedom". We can't know for sure what software they're actually running and users don't have any freedom in using their service since that is entirely at their pleasure. For something to be free the thing has to be able to move beyond its present controller.
Codeberg has its own issues with anti-scraping measures often impacting (not just annoying) users or some deep ideological bias; and I'm not talking about LLM/open source, which I agree with.
i'll never defend microsoft, but it irks me to hear the phrase 'free as in freedom' used specifically to describe a service that , as an advertised benefit, restricts user freedom.
no thanks.
I don't have to be pro or against AI to avoid signing a deal with (another) devil.
p.s. : "my code won't be training AI if im on codeberg" is just straight delusion, regardless of their policies.
I love and use Forgejo which they are both running and now developing within their ecosystem but development is so slow and such basic things are missing that one cannot help but think perhaps they should get some LLM assistance. At this rate they are very likely to end up a historical footnote.
One example Projects (boards). I would not necessarily call it missing for my own use case but it might as well not exist in its current state.
Label management is a mess too with almost no ease of use features. Lots of clicking when prior art already exists elsewhere.
Federation is another thing that's supposed to happen but isn't.
Action runners I like actually but their web UI is really confusing.
API is missing a lot of endpoints too. (Pretty straight forward for AI agents to work on as "good first issue")
No machine users or management of the same via Administrator. If you want to have a bot (not necessarily AI) you need to create a user with a hard to guess password and manage it by actually logging in as that user.
Another user will probably have their own set of things that matter to them.
As I've already said above, I do use it and love it but sort of half hoped Gitea would have proceeded farther since Forgejo forked away. It hasn't.
OneDev looks much more capable on surface to me today but I have not used it extensively yet and it has some features not included in the self hosted version. Not as much as open core projects but some.
I'm studying a master using Claude, for instante, and it's a breeze.
I can understand every problem, ask and explanation for every paragraph evenreorganizae a clearly suboptimal PDF notes from my teacher's class or Latex.
I use a stainless steel moka pot, which I find very efficient (a minute or less to get coffee ready), healthy and eco-friendly. I hate capsules because of the waste they create once used and the energy that goes into making them.
From what I've read, it's true that the oils aren't filtered out as well as with paper filters, but their contribution is minimal.
Yeah me too. Stovetop pot on induction stove. Takes less space than a coffee machine, less waste, and the coffee comes out hotter. Oh, and I can make two cups at once.
Those pod machines are just all around crap. Kind of acceptable in an office kitchen, but really surprised anyone wants one at home.
Love them or hate them, from a customer's point of view AMD is the only company in a position to stand up to NVIDIA in the GPU market and force those sky-high prices down a bit. I'm eagerly waiting for their new models, now backed by a much more mature and well-thought-out ROCm.
I'm not joking. I'm running a Maxwell-era NVIDIA card, and on Linux it's one problem after another. I HATE NVIDIA's driver; it's one of the worst-made pieces of software I've ever seen.
The Maxwell architecture is 12 years old. Like, I'm not out here defending NVIDIA's Linux drivers, but I also think it's unreasonable to expect great support for extremely complicated hardware that's over a decade old.
I've personally had zero issues with NVIDIA on Linux on both Turing (20xx) and Blackwell (50xx).
I agree that it’s an old, outdated, and obsolete architecture... but it is perfectly well-preserved and meets my current and future needs. I look at it like a car or a motorcycle; it might be 10 years old and lack an ADAS safety system or a giant touchscreen, but it’s still functional—in other words, I get to choose when to replace it.
The problem with NVIDIA is that its driver is an incredibly delicate component of its GPUs. With each new generation, the company implements—via software, through micro-routines—features that it failed to deliver in the hardware itself. If the driver were open—meaning one that the community could develop and maintain—I could continue using my GPU without any issues. Intel and AMD support the Mesa graphics stack, with full functionality. Why doesn’t NVIDIA?
The answer is that it doesn't make any financial sense for a hardware company in the business of selling new hardware to spend any software development resources at all on products that are 12 years old. There's a very clear conflict of interest.
Your problem is with capitalism, and I'm with you, but what you should be pushing for is legislation that forces companies to publish detailed specs for obsolete hardware so open source enthusiasts can do the work. Call it an extended "right to repair" or whatever.
In the meantime, have you tried the Nouveau driver for your setup?
I will be eternally grateful to Windows 11 and its TPM 2.0 requirement, and to Windows 10 for its senseless end of updates.
ETERNALLY, you read that right. Because thanks to that, I finally dared to try Linux, openSUSE, Doom Emacs and KDE, and I'm IN LOVE with my PC all over again.
It’s a beautiful story. Having a business like this in the AI era is VERY inspiring. I see you as a small business, like the butcher or the fishmonger, selling good products to the public. Congratulations, and keep it up.
The problem with AI is that, this time, nobody was caught off guard. From small startups to tech giants, everyone bet heavily on it. There are very good open-weight Chinese models, and even a French miracle. There is no winner-take-all.
So models are already a commodity, and will remain one for years. Companies must balance capex with very long payback periods against strict austerity everywhere else.
> From first principles I think youre right, but also Anthropic is making something like 80% margins.
I'm skeptical that they are operating at anything but a loss. 80% margins on a specific product (API) might still not cover the losses on the other products (Free tier, subscriptions) and their expenses (operating costs, training, research, etc).
reply