I wanted a better way to code with ChatGPT, so I built SVI

I’ve been using ChatGPT for coding for quite a while, but after some time, I realized that I was spending more and more time preparing prompts.
Every new file meant gathering context, deciding which existing files to include, adding project-specific instructions, copying everything into ChatGPT, and hoping I hadn’t forgotten something important.
So I built SVI (Structured Vibe Coding).

The idea is simple: instead of rebuilding prompts every time, SVI lets you define them once as Markdown-based .svi.md files. It then assembles the final prompt automatically from several pieces:
• instructions for the target file
• relevant code dependencies
• reusable prompt modules
• project configuration

The result is one complete prompt for a single output file.
If you don’t want to use an API, SVI now has a copy/paste workflow:

1. Run svi run -c
2. The complete prompt is copied to your clipboard.
3. Paste it into ChatGPT (or Claude, Gemini, etc.).
4. Copy the generated code back into SVI.

You keep the same structured workflow while using your favorite AI chat.
If you prefer automation, SVI also works with API keys from OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, and other providers supported by LLM.js. You simply bring your own API key.
Maybe you will find my tool useful, and looking forward to any comments.

GitHub: https://github.com/avrmsoft/svi