TinyCloud

World Builder is an AI assistant that lets anyone spin up a fully fledged World mini-app—no coding required. A user simply describes the idea in chat; the assistant turns that conversation into a concrete feature plan, scaffolds the entire project, and pushes a live deployment.

The magic ingredient is TinyCloud, a capability-based personal-data layer that lets mini-apps read and write user data in a privacy-preserving way and even run zero-knowledge checks across apps.

This means, for example, one mini-app can record a credit score while another only learns whether that score exceeds a threshold, unlocking richer cross-app experiences than today’s dapps, which are little more than front-ends for smart contracts.

Although demonstrated inside the World ecosystem, the same pattern can empower Farcaster frames, TON mini-apps on Telegram, or any platform that needs private, interoperable user data.

We began by forking an official World repository and grafting in TinyCloud’s SDK, giving the boilerplate app secure, shareable storage. On the front end, a lightweight Next.js chat interface collects user requirements; Railway hosts this interface so we could iterate quickly.

Once the user is satisfied with their app proposal, a backend Node service summarizes the chat and sends the spec to a dockerized claude-code instance. The app is build there as a git repo, and and turned into a zip file.

That artifact is piped back to Railway, whose build pipeline compiles and deploys the new mini-app to a fresh environment, yielding a public URL in minutes.

TinyCloud handles data capabilities, while Railway’s per-environment variables let us inject app-specific keys and endpoints.

For the hackathon we used a “one-shot” generation approach—faster to build, though less interactive than iterative scaffolding—and deferred our planned payment feature, which would let a verified World user mint one free app and pay one $WORLD token for each additional app.

Along the way we hacked together a TinyCloud shim to retrofit private data into a repo never designed for it, and scripted dynamic Railway stacks so every generated app spins up its own isolated Next.js instance on demand.

The result shows how AI plus an interoperable data layer can turn Web3 mini-apps into a truly composable ecosystem -- like a Chinese super app like WeChat