How this was built

Kafumu is written with an AI assistant, on plans made long before AI existed. Here's what that means.

Honestly: this is vibecoded

Most of Kafumu's code, its tests and its translations are written by an AI coding assistant (Claude), working in short loops: it builds a piece, runs the tests, puts it online, and I try it on my phone and say what's wrong. I decide what Kafumu is and does; the assistant does most of the typing. If that's a reason for you not to use it, that's fair, and it's your call. I'd rather tell you than have you find out.

The plans are much older than the AI

The ideas behind Kafumu were worked out years before AI assistants existed, by hand:

• olc-tools (since 2019): turning a location into a short text code, the #geo cells, and whenwhere, local messages by place.
• The Open Location Network whitepaper, and eolnpoc, its proof of concept: messages at a place, proof of work instead of accounts. How it works →
• lokumo, an earlier go at "what's around me", whose name and hosting Kafumu inherited.
• Language projects like esperanto-kurso.net and holywritings.net, where the account system and the cost habits come from.

Why build it this way

I'm one person. Working all of this out by hand is not viable for me without burning out; with an assistant doing the heavy lifting, it is. That's the whole reason.

What that means for you

• Translations are machine-made. If something reads oddly in your language, tell me and it gets fixed.
• Privacy doesn't rest on trusting the code blindly: private things stay on your device by design, the server holds only what it needs (see Privacy), and the code is open for anyone to read.
• Everything is tested before it goes online, including a two-browser test of connecting, chatting and syncing, run against the live site.

— Joop