Introduction

KERN is a compact language, compiler, and semantic review toolchain for AI-assisted software. You can write .kern yourself, or ask an LLM to draft it, then compile and review the result.

How it works

KERN is small enough for humans to read and regular enough for LLMs to generate reliably. The same source can describe UI, routes, types, tests, MCP tools, and review intent without forcing the model to emit every framework detail up front.

The practical loop is simple: describe the change, inspect the .kern, compile to your target, run review, and iterate with evidence instead of a wall of generated code.

What you get

Why these docs exist

If the AI writes KERN, why do you need docs? Three reasons:

• Understand what the AI generated — Read the .kern output, know what each node does, verify it's correct.
• Guide the AI better — The more you know about KERN's capabilities, the better prompts you write. "Add a state machine for the checkout flow" beats "make the checkout work."
• Trust the output — Understand the compiler, the review rules, the config options. Ship with confidence.

What KERN looks like

You'll see this in AI output. Each line is a node with props and styles:

page name=Hello
  metadata title="Hello World"

  col {ai:center,p:80,gap:16}
    text value="Hello World" tag=h1 {fs:48,fw:800,c:#fff}
    text value="Built with KERN" tag=p {fs:18,c:#a1a1aa}
    button text="Get Started" to=start {bg:#f97316,c:#fff,br:8,p:16}

The Syntax Reference explains every piece. The point is not to memorize every node; it is to know enough to guide, review, and correct generated .kern.

Get started

npm install -g @kernlang/cli

Choose your path

Deep reference