VAAC - Vibe as a Code

Keep AI-generated code separate from human code.

Versioned. Reviewable. Reproducible.

The Problem

AI-generated code is everywhere, but it's a mess:

• ChatGPT snippets copy-pasted inline
• Copilot suggestions scattered across files
• No way to track what's AI vs human
• Impossible to audit or review

VAAC fixes this. All AI-generated code lives in one place: the .ai/ directory.

How It Works

your-project/
├── src/
│   └── utils.ts           ← You write the prompt here
└── .ai/
    └── slugify-01.ts      ← AI generates code here

1. Annotate your function:

import { Ai } from '@gace/vaac';

@Ai({
  id: 'slugify-01',
  prompt: 'Convert text to URL-friendly slug. Lowercase, replace spaces with dashes, remove special characters.'
})
export function slugify(text: string): string {}

2. Run your build:

npm run build

3. VAAC generates .ai/slugify-01.ts:

export function slugify(text: string): string {
  return text
    .toLowerCase()
    .trim()
    .replace(/[^\w\s-]/g, '')
    .replace(/\s+/g, '-');
}

4. At build time, your function body is replaced with the generated code.

Your source stays clean. AI code stays separate.

Installation

npm install @gace/vaac

Add to your vite.config.ts:

import { defineConfig } from 'vite';
import { vaac } from '@gace/vaac/vite';
import 'dotenv/config';

export default defineConfig({
  plugins: [
    vaac({
      model: 'openai/gpt-4-turbo',
      verbose: true,        // Optional: enable debug logging
      // apiKey: '...',     // Optional: auto-detects from env vars
    }),
  ],
});

Create a .env file with your API key:

# OpenRouter (recommended - access to multiple models)
OPENROUTER_API_KEY=sk-or-...

# Or use OpenAI directly
# OPENAI_API_KEY=sk-...

Philosophy

As AI-generated code becomes ubiquitous, mixing it with human-written code creates chaos - hard to track, hard to audit, hard to manage.

VAAC isolates AI code. Every generated function lives in .ai/, committed to Git, reviewed like any other code. Your human code stays clean.

Key Features

| Feature | Description | |---------|-------------| | Separation | AI code in .ai/, human code everywhere else | | Versioned | Generated files are committed to Git | | Reproducible | Same commit = same build output | | CI-safe | Build fails if .ai/ files are missing - no surprise AI calls | | Zero runtime | Generated code is compiled in, no runtime dependency | | Editable | Don't like the AI output? Edit .ai/ files directly |

Developer Flow

| Action | How | |--------|-----| | Generate code | Run npm run build - missing .ai/ files are created | | Review AI code | Open .ai/[id].ts - it's just TypeScript | | Edit AI code | Modify .ai/[id].ts directly - your changes persist | | Regenerate | Delete .ai/[id].ts and rebuild | | Commit | Add both source and .ai/ files to Git |

Configuration

vaac({
  // AI model (default: gpt-4-turbo)
  model: 'openai/gpt-4-turbo',
  
  // Output directory (default: .ai)
  outputDir: '.ai',
  
  // Enable debug logging (default: false)
  verbose: true,
  
  // API key - auto-detects from env vars
  // Priority: OPENROUTER_API_KEY > OPENAI_API_KEY
  // Or pass explicitly:
  // apiKey: process.env.OPENROUTER_API_KEY,
  
  // CI mode: error if files missing instead of generating
  ciMode: process.env.CI === 'true',
})

License

MIT

Built with the belief that AI and human code shouldn't mix.