We’re building a factory that turns prompts into real apps you can run.
Today the factory takes a natural-language prompt and produces a working Next.js web app with real persistence (libSQL), full CRUD pages per entity, and functional smoke verification — every job marked “pass” means the bridge POSTed a row to the live server and got it back. See /factory/proof for the live list of every passing run with a downloadable .tar.gz. Desktop installers and mobile APKs are on the roadmap, not shipped.
Why we built this
The promise of AI app generation is that anyone can describe what they need and get working software. That promise is largely kept — for slick web demos. Drop a prompt, get a React landing page in 10 minutes. Click any link and you 404.
We wanted a factory where the output is the actual thing — with real database tables, real CRUD forms, real persistence — not a scaffold with mock arrays. Where every passing run produces a downloadable artifact you can extract, npm install, and run.
The bigger vision — Windows installers, mobile APKs, dedicated bridges per tier — is roadmap, not shipped today. The Windows bridge architecture is in factory/, but no desktop EXE has been built end-to-end on prod yet. The mobile packager exists as a stub. We’ll mark them shipped when they verifiably are.
We built this source-available, so customers can verify what we say. Every honest claim on our pages links to either a file path you can read or a row on /factory/proof you can click. If we say a feature is shipped, you can see it run. If we say roadmap, you’ll see it labeled that way.
How we work
We ship daily
Public changelog at /factory/changelog. Multiple PRs land most days. Customer-visible work is logged, fix-typo commits are not.
We say "partial" honestly
Our comparison matrix marks features as partial when code exists but isn't proven end-to-end. We don't fake "shipped".
We optimize for the customer who reads code
Source-available on GitHub. Every architectural claim is verifiable. Procurement teams love this.
AI is a co-author, not the only one
We use AI heavily in our own dev loop. Every commit credits both the human and the model. We pair-program with Claude.
No vendor lock-in
Artifacts you generate are yours — downloadable, exportable, redeployable elsewhere. Audit logs are JSONL-streamable.
Customer-funded, not VC-funded
We’re bootstrapped. Roadmap is driven by paying customers, not growth metrics. Smaller, sharper team.
What we don’t do
- ✕We don't train models on your prompts.
- ✕We don't sell your data — not now, not ever.
- ✕We don't lock you into our runtime. Artifacts run anywhere.
- ✕We don't add fake features to the comparison matrix.
- ✕We don't hide the code. The whole factory is on GitHub.
- ✕We don't ghost on support. Email and we respond within 24h.