Fluxer

Fluxer is a Discord alternative that makes use of AI in it's development. We can't recommend against it enough, please do not use and/or support it. The fact that this post has reason to exist is disappointing, but we would rather people be aware of the situation.

AI Use

Fluxer's heavy use of AI isn't even fully disclosed to the public, as their Git history prior to their first release as been purged. In this post, Kraft claims it's "common practice", but that's usually in the case of propritary software going public, or if credentials are left in the source code by mistake. It's not often used because you "...didn't want to make 3,000+ messy commits part of the public record." This is an attempt to conceal the history of what is, and was presumably always intended to be, open source software. It is not normal within FOSS and especially not for the reason given. While the linked post is correct that maintainers should not be harassed over LLM/AI use, that does not mean it should be acceptable in programming or elsewhere. Kraft promoted Fluxer's CLA in that same post, so let's tackle it next.

The CLA

Fluxer's CLA gave them the ability to go proprietary in the future if they so desired. While eventually removed due to community backlash, the following was also stated prior to the CLA's removal: "Finally, we can offer commercial licences to companies that want to run Fluxer internally without being bound by the AGPLv3 copyleft terms. This is enabled via a contributor-friendly CLA, but it doesn't create a separate 'enterprise edition'. It's still the same Fluxer software everyone else uses." This is PR speak for "we can make it proprietary for those willing to pay".

Transparency

You might have noticed that the referenced links lead to versions of the post preserved by The Wayback Machine. This is for a good reason, as the post has been edited at least four times since it's original date of publishing, which still sits erroneously atop the post. There was no indication that any changes had taken place. This is problematic in its own right, but even more so when we consider what was silently revised. Firstly, all references to the (admittedly now defunct) CLA were removed. Secondly, the AI policy was changed to make it, as some would describe, "more reasonable". Thirdly, the section in which AI is discussed was removed entirely. Afterwards, there was a brief period where Fluxer asked AI use to be disclosed. While this still allowed it to be used in development, it's mildly better than not requiring any disclosure at all, which is where the project stood for a time. As of writing this, however, it seems they've adopted a more unclear policy, now stating that "If an LLM suggests code, wording, a defect or a solution, the contributor MUST verify the underlying information independently," and that "The contributor MUST keep the LLM output private. The contributor MUST NOT publish it, include it in a submission or require another person to read or evaluate it." This seems to be a step backwards, not having a clear policy on when AI is allowed. They do at least clearly state that AI is allowed to be used for various tasks including ingestion.

In addition, the URL for the blog post has changed twice, which makes it much harder for anyone to find previous versions of it on the Wayback Machine. It was originally hosted at https://blog.fluxer.app/how-i-built-fluxer-a-discord-like-chat-app/, then https://fluxer.app/blog/how-i-built-fluxer-a-discord-like-chat-app, and now https://fluxer.app/blog/how-i-built-fluxer, at the time of writing. All of these previous URLs re-dirrect to the newest version, but it makes discovering the old revisions harder still.

The LoC problem

Harmony, at time of writing, is made up of 38,133 lines of TypeScript. Stoat is 46,433 lines of Rust, OldCord is 23,554 lines of JavaScript, and in comparison, Fluxer contains 882,379 lines of Typescript, 189,748 lines of Rust, and 103,355 lines of Erlang. For a larger example: Matrix is a far more complex project that's been around for longer and has had more cooks in it's digital kitchen. Its Synapse server is only 381,006 lines long. Fluxer contains 3 times the amount of code as Synapse (Matrix), 25 times as much as Stoat, and 31 times the amount that Harmony contains. The first public commit to Fluxer is only around 8 months old and it alone has 302,671 lines of TypeScript, 52,004 of Go, and 43,451 of Gleam. It is twice the size of modern Matrix and far exceeds Stoat and Harmony, both of which have real histories made by real contributors over time. Additionally, all this code was written by one person, over a vague period of time that people can't seem to agree on. The project gained about 800,000 lines of code in 9 months. For perspective, Linux gains about 100k lines of code every month (40mil lines in total / 34 years of existing / 12 months in a year) on average, and Fluxer nearly kept pace. This is a clear sign of AI slop or, at the very least, an absolute mess of a codebase.

Uploading Files to Third parties

Fluxer uploads full file contents to third parties, something I and many others are rightfully against. We want a platform better than Discord, not one that replicates their mistakes and disrespects it's users. That's why people are leaving it behind, and what's the point in going elsewhere if it's doomed just the same?

Federation

They initially claimed "Federation is in development", but then later said that "There is no plan or date for federation," which implies that the earlier claim was incorrect, or at least deceptive.

Using FOMO to Make Money

Early in it's existence, Fluxer took advantage of the fear of missing out FOMO to sell people their "visionary tier", which was and still is a rather large ask for a project that was not even a month old at the time. It sadly worked, and people bought into it.

"Made in Sweden"

It might be true, it might not be. We're inclined to believe it, but it's deceptive either way. Fluxer claims to be made in Sweden, but you might notice they don't claim to be hosted there, because they're not. They also now include a Swedish flag on their homepage next to their logo. If we look at their API (which is currently at https://web.canary.fluxer.app/api/v1), the current IP of 144.202.15.48 is located in the USA. While it's near impossible to find privacy laws that don't need serious reform, the USA's are uniquely concerning and only getting worse. It is partially why Discord has gotten as bad as it is.

And more...

This is just a summary of what I can recall off the top of my head, but it shows a clear pattern. Fluxer cannot be trusted, and are willing to forgo the possibility of a coherent and sustainable platform for profit, just like Discord has. If you know of anything I missed, please email me with proof, and I'll add it to this post.

Thank you for reading.

Researched & written by Mathium05

Proofread & edited by cloudskater

In case you're interested, we keep a git history of our blog posts, making it easy for you to tell when we edit them.