Website Generator

The Website Generator project investigates how language models can serve as full-stack UI engineers—from prompt to component tree, from data schema to deployment-ready code.

Inspired by tools like V0, Bolt, and Claude Code, we’re building our own open foundation for AI-native front-end development, driven by declarative design, component libraries, and developer-in-the-loop feedback cycles.

What We're Building

At the heart of the project is an AI-to-UI compiler—a model-driven system that takes natural language prompts and produces structured websites using modern stacks like:

• Next.js / React
• Tailwind CSS / ShadCN
• JSON Schemas → UI forms
• GraphQL / REST hooks

Output is not just static HTML—it’s editable, extensible, and deployable.

Core Technologies

• Code LLMs: Claude 3, GPT-4 Code Interpreter, DeepSeek Coder
• Abstract UI Language (AUL): Our internal prompt-to-UI intermediate format
• Live Editor Loop: LLM + diff patcher + file context retention
• Design Tokens: Standardized style guides injected into prompt context

Features in Development

• Prompt-to-layout: Describe a homepage → get a responsive, styled layout
• Prompt-to-form: Provide a schema or text → generate full input components
• Component rewrites: "Make this dark mode", "Add animations", etc.
• Local preview & deploy: Generate → preview → push to endpoint in minutes
• Multi-file context: Inject custom data models or utility functions as context

Open Source Philosophy

Unlike closed platforms, our Website Generator is built around transparency and reusability:

• All outputs are clean, readable code—no obfuscation
• We use and contribute to open libraries like tailwind, radix, shadcn, valtio, and more
• Our internal UI builder is being modularized for future release under a permissive license
• Outputs are framework-agnostic, not locked to a vendor

Example Prompt

"Build a responsive SaaS landing page with a pricing table, customer logos, and a contact form."

Output:

• /components/Pricing.tsx
• /components/LogoGrid.tsx
• /pages/index.tsx
• Tailwind utility classes + layout grid
• Markdown-powered copy blocks

TensorOne Stack Integration

All generation and previews are powered by:

• TensorOne Serverless Endpoints for LLM inference
• TensorOne File Storage for snapshot versioning
• Cluster rendering using headless Chrome for screenshots and CI snapshots
• CLI flow:

tensoronecli project create --type webgen
tensoronecli project dev

What’s Next

• Design-to-code: Upload Figma or PNG → generate React components
• Data-aware dashboards: Auto-generate UIs from mock APIs or schema introspection
• Multi-agent generation: Writer, Designer, and Engineer agents collaborating on web projects
• Open Playground: Editable live coding environment with diff visualization

The future of web development isn’t no-code—it’s AI-native code.
We’re not replacing developers. We’re accelerating them—from idea to interface.