Linuxdownload.org, vibecoded end-to-end

A new toy project

Meet linuxdownload.org, my latest weekend toy. The idea is simple: make it easier for someone curious about Linux to actually get started. The site has side-by-side distro comparisons, a “which distro fits me” quiz, an app-equivalent finder (think: “what do I use instead of Photoshop?”), gaming readiness scores, post-install setup generators, and even a Windows 10 migration guide for those finally jumping ship now that Win10 is EOL. Over 20 distros covered so far, from the obvious (Ubuntu, Debian, Fedora) to the more exotic (CachyOS, Bazzite, NixOS, Tails).

Built exclusively with AI tools

What’s interesting about it, at least to me: I didn’t write a single line of code or push a single pixel by hand. Every part of the project, design, frontend, backend, scraper, infra, was created with AI tools.

For the design and overall vibe, I used Google Stitch to mock up the layout and color palette. Surprisingly, the output didn’t have that generic “AI-generated UI” look that everything seems to default to these days.

For the code, I leaned on qwen3-coder running locally on the Ryzen AI machine for the cheap, fast iteration loop, and Claude Code for the heavier lifting. The stack is FastAPI + SQLite on the backend, Caddy in front, and a Python scraper that auto-detects the latest ISOs and their SHA sums for each distro on a schedule, so the download links and checksums never go stale. All of it agent-written.

Is “vibecoding” actually viable?

For a toy project like this? Absolutely. From zero to a deployed, working site in a couple of evenings. The catch: you still need to know what you want, when something is off, and how to wire the pieces together. The AI is fast at typing; it’s still you doing the thinking.

Anyway, give it a spin: linuxdownload.org. Feedback welcome.