The motivation: AI app builders got good at generating apps and stayed bad at letting you leave. Frontend code you can always lift out. The backend is the hostage — their runtime, their DB, their deploy target.

How it works: you wire blocks on a canvas (HTTP routes, tables, auth, queues, cron, multi-step pipelines) into a graph. That graph is an IR, and the compiler lowers it to a conventional Node project — typed Express handlers, Drizzle schema and migrations, Zod validators derived from the table definitions, `postgres` for the driver. Nothing is interpreted at runtime; the graph isn't shipped, it's compiled away.

The test I hold it to: the exported package.json lists express, drizzle-orm, postgres, zod, dotenv, jsonwebtoken. No package of mine. Delete your Wyro account and the app keeps running.

Databases work the same way — connect your own Supabase and provisioning happens in your org, on your bill, over the Management API. I hold the canvas, not your infrastructure.

Honest state of things: it's early and solo-built. The generated code is the thing I'm most confident about and the thing I most want torn apart. Rough edges are in the surrounding product, not the compiler.

Free tier, no card. The useful thing you can do in five minutes is export a project and read the diff-able code before deciding whether you trust it.

https://wyro.vercel.app

Would especially like feedback from anyone who has migrated off a hosted builder — what broke, and what you wished the export had included.