Ask an AI for a web app and you usually get one of two things: a single HTML file that demos fine but isn't a project, or a wall of code fragments you're expected to stitch together yourself and hope they compile.
Aii Cloud now does the third thing — the one you actually wanted. It writes the whole project, real structure and all, then builds and runs it in your browser so you can use the app before a single file touches your machine.
Here's the proof, live right now:
That's a complete machine-learning lab: generate a dataset (XOR, circles, spirals), configure a network layer by layer, hit Start Training, and watch the loss curves fall and the decision boundary take shape — the model trains in your browser tab, on the spot. Click any point on the plot and the trained network predicts its class. There's even a plain-English explainer for every concept on the page.
It was built from one prompt in Aii Cloud's Agent mode — and you don't have to take our word for it. The conversation that built it is public: read the exact prompt and watch the build unfold.
A real project, not a snippet
The prompt asked for an interactive playground that trains a small binary classifier entirely in the browser with TensorFlow.js and verifies generalization on a held-out test set — a multi-file Vite-style React + TypeScript app, no backend. Agent mode planned the work, researched the current TensorFlow.js APIs on the web, and then wrote the project the way an engineer would:
index.html · package.json · vite.config.ts · tsconfig.json
src/main.tsx · src/App.tsx · src/index.css · src/types.ts
src/utils/data.ts · model.ts · metrics.ts
src/components/ — 8 panels, from DatasetPanel to ExplanationPanelNineteen files, typed end to end, with the training logic commented and the UI split into components you'd be happy to inherit. Every file lands in the workspace beside the chat — browse the tree, open anything in the built-in editor, tweak a line and save.
It runs before you download anything
This is the new part. The workspace Preview doesn't show you a mockup of the code — it builds the project, right in the browser: TypeScript compiles, the npm dependencies your package.json declares (TensorFlow.js, Recharts, React) resolve automatically at their pinned versions, styles apply, and the app mounts.
So thirty seconds after the last file streams in, you're not reading code and imagining it working — you're clicking Generate Dataset and training a model inside the chat. If something's off, tell Aii in the next message; new files join the same project and edits version cleanly, so the preview just keeps up.
Take it home, or put it on the web
When it's yours to keep, two buttons:
- Download grabs the whole project as a zip —
npm install && npm run devand it runs locally, because it was a real Vite project all along. - Publish puts it at a named address. This demo lives at
nn-demo.site.aiicomputer.xyz— a free named Site, same as any page Aii builds for you. Share it, update it, take it down — the address is yours.
Try it yourself
Open cloud.aiiware.com, switch to Agent mode, and ask for something real:
Build a Vite + React + TypeScript app that visualizes sorting algorithms step by step, with speed controls and plain-English explanations of each algorithm. No backend.
Watch the plan appear, the files stream in, and the preview come alive. Then publish it and send someone the link.
Still free to start, still no API keys required.