kun

kun (كُن) — "Be." An intent-driven code generator.

You state intent. Kun translates it into real code.

What is kun?

kun is a CLI tool that generates source code from .kun requirement files using LLMs. You describe what a file should do, and kun produces a working first draft.

SearchBar.vue.kun   →   kun build   →   SearchBar.vue

The filename encodes the output. Everything before .kun is the file that gets generated.

Install

npm install -g @seyyidt/kun-js
# or
npm install --save-dev @seyyidt/kun-js

Requires Node.js >= 20.

Quick start

1. Configure your LLM provider

Create a kun.config.js (or kun.config.ts) in your project root:

// kun.config.js
import "dotenv/config";

export default {
  provider: "anthropic",
  model: "claude-sonnet-4-20250514",
  apiKey: process.env.ANTHROPIC_API_KEY,
};

2. Write a .kun file

@target vue3

A button that copies text to clipboard.
Shows "Copied!" for two seconds after click.
Uses no external libraries.

3. Generate

kun build CopyButton.vue.kun    # generates CopyButton.vue
kun build                       # builds all .kun files in the project
kun build --dry-run             # preview the prompt without calling the API

.kun file format

Plain, structured natural language. No strict syntax — clarity is the only rule.

@target    vue3
@component SearchBar

intent:
  A search input used across the app header.
  Debounces input and emits a search event.

props:
  placeholder: string (default: "Search…")
  debounceMs:  number (default: 300)

emits:
  search(query: string)

behavior:
  on input  → debounce, then emit search
  on Escape → clear input and emit empty string

constraints:
  - no external libraries
  - fully keyboard accessible

Write what, not how. The LLM decides the implementation.

Providers

kun is provider-agnostic. Built-in support for:

| Provider | Config | |---|---| | Anthropic (Claude) | provider: "anthropic" | | OpenAI | provider: "openai" | | Any OpenAI-compatible endpoint | provider: "custom", baseUrl: "..." |

// OpenAI
export default {
  provider: "openai",
  model: "gpt-4o",
  apiKey: process.env.OPENAI_API_KEY,
};

// Local / Ollama / vLLM / LM Studio
export default {
  provider: "custom",
  model: "llama-3.3-70b",
  baseUrl: "http://localhost:11434/v1",
};

How it works

kun build reads a .kun file, computes the git diff against its last committed state, and sends both the full requirements and the diff to an LLM. If a source file already exists, the LLM patches it. If not, it generates a new one.

The git sync contract

When you commit both a .kun file and its generated source file together, they are declared in sync. After that, you own the generated code — edit it freely. The .kun file remains as documentation of intent.

License

MIT