I kept running into the same problem with AI-generated frontend code: most tokens go to boilerplate, and the AI can never pick a consistent pattern. So I built a language where there's only one way to write things.

0x is indentation-based (think Python), declarative, and compiles to React JSX, Vue 3 SFC, or Svelte 5. A counter component is 18 lines in 0x vs 96 in production React.

page Counter: state count: int = 0 fn increment(): count += 1 layout col gap=16 padding=24 center: text "{count}" size=4xl color=cyan button "+1" style=primary -> increment() The compiler is ~3K lines of TypeScript, zero dependencies. Pipeline: Lexer → Parser → AST → CodeGen (one pass per target). It handles state, derived values, typed variables, functions, flexbox layouts, control flow (if/elif/else, each, match), lifecycle hooks, API calls with loading/error states.

I chose indentation-based syntax because it's the most token-efficient structure I could find. No curly braces, no semicolons, no JSX closing tags, no import boilerplate. For an LLM, fewer structural decisions = fewer hallucinations.

There's a built-in MCP server so Claude and Cursor can compile inline. Also works as a library:

import { compile } from '0x-lang/compiler'; const result = compile(source, { target: 'react' }); I'm curious about two things:

Is "designed for AI" a real market, or is this too niche?

For folks who've built compilers — any obvious mistakes you see in the architecture?

Website: https://0xlang.com GitHub: https://github.com/hankimis/0x-lang npm: https://www.npmjs.com/package/0x-lang