Every app here was built by a machine.
Last updated 2026-07-16
Almost none of the apps on this site were written by hand. They're the output of a project factory — an autonomous AI system I built that ideates, specs, codes, tests, ships, and even names and markets small apps around the clock, whether I'm at my desk or asleep.
Why build a factory?
I pay $200 a month for a Claude Max subscription — a flat-rate plan for the AI model I use to write software. Flat-rate is the operative word: it costs the same whether I use it constantly or not at all. Real life means some days it sits idle. Some weeks, too. That idle, already-paid-for capacity bugged me.
The factory is my answer. It makes sure every drop of that subscription gets used building something. It runs 24/7 on a small server and paces itself against the plan's usage limits: when usage is behind the month's pace it works harder — more runs, a stronger model — and when it's caught up, it eases off. A subscription that used to go partly unused now produces the site you're looking at.
The loop
A Python orchestrator runs on systemd timers and reconstructs the world from disk and GitHub on every tick — there's no long-running brain. Each project moves through the same pipeline, one bounded AI work session at a time:
- Ideate. An ideation pass generates candidate ideas, each with the real search phrases it would be found by. The orchestrator then probes those phrases against Google Autocomplete, Reddit, and Hacker News and computes a demand score. Priority is a weighted composite — model-scored axes plus measured demand plus monetization fit — and the model never gets to self-score the measured parts.
- Scope. The top idea gets a public GitHub repo, a vision doc (the problem, the user, what "done" means), and an art-direction brief written before any UI code exists.
- Build. Headless coding sessions run in a memory-capped sandbox with an activity watchdog — a run is killed only when provably hung (no commits, no file writes, no CPU). Work lands as small atomic commits, pushed continuously, so the repo is public while it's being written, not after.
- QA. Fresh sessions with no memory of the build run the test suite, fix, and refactor. The orchestrator measures commits and test results itself via git rather than trusting the model's self-report, and nothing promotes with red tests.
- Ship gates. A stronger model reviews the finished project — code quality, a design gate that can reject it back into another round, a secrets scan, and an authorship check. The same pass picks the final product name (the repo is renamed to match) and writes the copy around the measured search phrases.
- Publish. The build is copied into the showcase repo under its slug, the registry tile is upserted, and the sitemap, Atom feed, and structured data are regenerated. A git push triggers CI, which deploys to the CDN; search engines get pinged and the launch is announced automatically.
- Monetize, sometimes. The site runs ads, pages carry query-targeted structured data, and there are gated experiments (payments, app-store wrapping) that stay off until explicitly armed. This is the least developed part — see the FAQ for honest numbers.
- Revisit. A freshness loop re-measures demand for shipped apps, proposes new search phrases, refreshes copy, and re-publishes.
Where I fit in
I built and maintain the factory itself — the orchestrator, the quality gates, the pacing logic — and I fix it when it breaks. When I have an idea I want to exist, I queue it, and queued ideas jump ahead of everything the factory would pick on its own. When I'm not using the subscription and there's nothing waiting, it doesn't idle: it invents its own next project, from ideation through spec, QA, publishing, and monetization.
FAQ
Did a human write these apps?
Mostly no. The AI wrote the code, the tests, the copy, and usually picked the name. I wrote the factory — the system that does all of that — and I take responsibility for what it ships.
Are the apps actually good?
Some are genuinely useful, some are fun toys, and a few are duds. Every project ships only after its tests pass and an AI review gate approves it, but no human hand-checks every line. If something's broken, open an issue on the app's repo and it gets fixed.
Is this site making money?
Barely, so far. There are ads (Google AdSense) and small monetization experiments. The subscription powering the factory costs $200/month, and as of mid-2026 the site earns nowhere near that. The factory exists because the subscription was already paid for — the apps are the return on capacity that would otherwise evaporate.
Is the code public?
Yes. Every app is a public, MIT-licensed repository on github.com/ctkrug with its full commit history.
The commits are under your name — isn't that misleading?
The commits are authored under my GitHub account because I own and answer for the work, and this page is the disclosure that an AI produced most of it.
Are these apps safe to use?
They're static, client-side pages: no accounts, no logins, and your data stays in your browser unless an app's own page says otherwise. The privacy policy covers the site's analytics and ads.
What actually runs the factory?
A Python orchestrator on a small Linux server, driving Claude (Anthropic's AI model) through the phased pipeline above — on the flat-rate subscription, not a metered API. That constraint is the whole point: the marginal cost of the next app is zero.
Can I suggest an idea?
Yes — send it via charliekrug.com/contact. Good fits go into the queue, and the queue gets built first.