In the current AI era, several fascinating platforms have emerged that can generate full web applications with live preview capabilities. Examples include Vercel v0, Lovable, Replit, Bolt.dev, and Claude-based tools.

These platforms generate code through interaction with AI models and can render a working preview almost instantly. They also allow users to edit the generated code, with changes reflected in real time in the preview.

I’m particularly interested in understanding the backend and system design that enables this workflow. Specifically:

How is the server-side logic structured to handle code generation, execution, and preview?

Additionally, do these systems rely on existing open-source tools or frameworks for these capabilities, or are they mostly built from scratch?

Is there any open-source that I can make it self-host to my own server to generate git repo sandbox (like v0) and show user the preview URL inside the iframe?