Is Fluxer vibe-coded?
No, Fluxer isn't vibe-coded. A person who understands the code writes every code change and checks it before it ships, and anyone can look through the source code and its history on GitHub. Andrej Karpathy coined the term vibe coding in February 2025 and described it as a way to "forget that the code even exists." It means letting an AI tool write software and shipping it without reading or understanding the code.
Who builds Fluxer?
Fluxer is a chat app made by Fluxer Platform AB, a small Swedish company. Two product engineers build it with help from a few contributors. One of them works on the iOS and Android apps. The other is Hampus Kraft, the founder and lead developer, who works on Fluxer full-time and builds the backend, the web app and the desktop app. He started building it in high school, opened the first private alpha to testers in July 2023 and wrote the whole story down in How I built Fluxer. He is close to finishing a BSc in Computer Engineering at KTH Royal Institute of Technology and has built and run large production systems for years.
I don't think vibe coding is cool, and I don't condone slop or unethical use of LLMs.
Does Fluxer train AI on my messages?
No. Section 3.1 of our terms of service is our promise about AI training and your content, and section 5 of our privacy policy lists the automated checks that run on your messages, files and calls.
Why are Fluxer's line counts so big?
Most of the lines that change in Fluxer's repository are translations. From August 22 to September 28, 2026, git counted 1.10 million lines added and about 980,000 deleted, a net gain of about 121,000. Translation files were 53.5% of the changed lines, tests were 12.3% and application code was 22.8%.
Fluxer is available in 34 languages and regional variants, and each sentence in the app takes about five lines in every one of their translation files. One new sentence adds about 170 lines, and rewording one can make git count all 170 as deleted and added again. The app's 8,000 or so sentences come to more than 1.3 million lines of translations. New sentences start as machine-translated drafts, and people review and improve them on Fluxer's translation site.
Some of the biggest commits are months of work shipped at once. Spam waves, attackers posting child sexual abuse material, DDoS attacks and rushes of new people after news about Discord kept pulling Hampus onto the most urgent problem. Each time, the work waiting to ship went out of date and had to be updated again, like Sisyphus's boulder rolling back down the hill. Once things calmed down, it shipped in a few large batches.
Fluxer's public history starts with one commit of nearly 1.5 million lines, dated January 1, 2026. About 800,000 of those lines are translations, and the rest is years of work from before the code was public. Starting the public history with a single commit is common practice when private code is open sourced, because a private history can contain passwords, keys and other details that shouldn't be published.
What Hampus doesn't condone
- Shipping code that nobody has read or understood. That's vibe coding, and it's why a person who understands the code writes and checks every change to Fluxer.
- Flooding open source projects and communities with slop. Our anti-slop policy keeps AI-made contributions out of Fluxer's code.
- Training AI on people's messages, files or personal data without their consent, or selling that data to AI companies. Section 3.1 of our terms of service says Fluxer never trains AI on your content and never sells or licenses it for anyone else's use.
- Making sexual images of real people without their consent. Section 1 of our community guidelines bans sharing intimate material without consent, including realistic synthetic imagery.
- Making any sexual image of a child. Section 4 bans it in every form, AI-generated content included, and breaking it means permanent account termination and a report to law enforcement where the law requires it.
- Impersonating someone or faking evidence of something that never happened. Section 7 bans using AI-generated media to deceive or impersonate anyone, or to pass it off as real evidence.
- Spam, fake accounts and fake engagement. Section 6 bans spam, fake or deceptive accounts and artificially inflated numbers.
Section 7 still allows AI-generated media for creative, fictional, educational and satirical use when it follows the rest of the guidelines. Allowed doesn't mean tasteful, and individual communities can set stricter rules.
Our anti-slop policy
People outside the Fluxer team can suggest changes to Fluxer's main repository on GitHub. Our anti-slop policy lets them use an LLM privately to research and plan their work. They may not submit code, text, images, audio or video made by an LLM or other generative AI, even after editing, and everything they submit must be their own work. The one exception is a disclosed machine translation of their own writing.
How many lines did this page add?
This page is in 34 languages and regional variants. The English version is 54 lines long. Its 32 translations, plus its title, summary and menu link in every language, added 2,472 lines to git, about 46 times the English page. Each translation started as a machine-translated draft, the same way the app's sentences do.