People ask us this all the time. You type "build me a meal planner," and a few seconds later there's a working app on your screen. What's going on behind the scenes?
Let's walk through it.
It starts with intent, not templates
When you say "a meal planner that suggests recipes based on what's in my fridge," we don't look up a meal planner template and fill in blanks. The AI reasons about what that description implies: you need a way to input ingredients, a recipe matching system, a display for results, probably a way to save favorites.
It builds the whole thing from scratch. Every time.
This matters because it means your app isn't a reskin of something generic. It's built around your specific description.
You see it running immediately
As soon as the code exists, it's running in a live preview. Click buttons. Fill in forms. It's the real thing.
This changes how you work with the AI. Instead of describing everything upfront and hoping for the best, you iterate. "Make the buttons bigger." "Add a dark mode." "Actually, put the results in a grid instead of a list."
Each change rebuilds the app. The conversation history means the AI remembers what you've already built, so refinements stack on top of each other.
Publish is the easy part
When you're happy, hit publish. We take your app and wrap it as a PWA: generate a manifest, write a service worker, resize your icon, pick up your theme color. It goes live at a URL like https://envelope.build/meal-planner.
That's a real link anyone can open, install, and use. No technical setup. Just publish and share.
What we're working on next
Right now we're adding modules: push notifications, a key-value database, a chatbot you can embed, voice input, AI-powered image generation. The idea is that your app can tap into these with a toggle, no code needed.
We're also getting better at understanding vague prompts. The more people build, the better the AI gets at knowing what "a booking app for my barber shop" actually means in practice.