Vibe Coding

Vibe coding lets you describe what you want to build in plain English and have the AI generate it for you — no template selection, no setup.

How It Works

On the Landing Page

If you're not signed in, you'll see a prompt input on the landing page. Type what you want to build and click Generate with AI.

• If your prompt is clear (e.g. "sorting algorithm in Python"), the AI selects a template and starts building
• If your prompt names a huge, unbounded idea (e.g. "build instagram"), the AI picks a specific, bounded starting slice and says so as its first message, rather than asking anything
• If your prompt is genuinely ambiguous (e.g. "build me a rocketship"), you're dropped into Plan mode automatically — see below
• Once ready, click Start Building — you'll be prompted to sign in with GitHub if you haven't already
• Your prompt is saved through the sign-in flow and will not be lost

On the Dashboard

If you're signed in, the same prompt input appears at the top of your dashboard. Type your prompt and submit — the editor opens with the right template, in a focused, chat-only view (no file tree or preview yet), and either starts building right away or asks a clarifying question first, depending on how clear the prompt was.

Attaching a screenshot or file

Want the app to look like something? Paste a screenshot straight into the prompt box (or the chat box inside an app) (⌘V / Ctrl+V), drag a file onto it, or use the + menu. Images, PDFs and text or code files work. The AI sees the attachment with your first message, so "I want it to look like this" means the image you attached.

If a file is still uploading — or failed to upload — when you press send, cdbx tells you instead of quietly building without it.

When a big build stops early

If a build finishes but your app's main file still shows the starter placeholder, cdbx notices and asks the AI to finish it. A large app can take more than one pass. If the AI runs out of steps part-way through, it continues by itself (up to three times) and you'll see short "Continue from where you stopped" messages in the chat. If it runs out of time instead, it stops and tells you: everything it wrote is kept, and you can send continue or ask for one piece at a time. A preview that still says "Bringing your ideas to life…" after a build means the AI hasn't written the main file yet — send continue.

Build vs. Plan

Next to the send button, a Build ▾ dropdown lets you switch to Plan mode before submitting, or switch it mid-conversation.

• Build (default) — writes files immediately.
• Plan — discusses first; no files are edited until you say go.

A genuinely ambiguous prompt switches this to Plan for you automatically, so it reads honestly from the very first message — you don't have to notice and flip it yourself.

What Plan mode looks like

In Plan mode, the AI asks whatever short questions it actually needs before building — as multiple-choice buttons, free text, or both — right inline in the chat. There's no fixed number of rounds: the AI decides whether one question is enough or a few more rounds are worth asking, the same way a person would. A round of questions shows one at a time, with arrows to page between them and a running count — answer or leave any of them blank to skip, then send once you reach the last one. The AI either asks another round or shows a Start building card summarising what it's about to do.

You're never stuck discussing — type "just build it" (or anything like it) at any point and the AI proposes to start building immediately, using reasonable assumptions for anything still unclear. You can also flip the toggle back to Build by hand at any time; the very next message gets full file-editing tools.

An app you're still planning shows as Plan in your list and stays there until something is actually built, so an unfinished one isn't lost — see Apps.

Template Selection

The AI picks the right template from your description:

You can leave any clarifying question blank and send — the AI treats it as skipped and works with what you already gave it.

In the Editor

When the editor opens:

• The AI Chat panel is automatically focused, full-width and centered — no file tree or preview yet, so the conversation is the whole screen
• Your prompt is sent as the first message
• The AI reads the template files and rewrites them to match your description (or asks a clarifying question first, in Plan mode)
• File changes are applied automatically (auto-accept is on by default)
• The moment the first file is actually written, the view animates into the normal editor — file tree, code, and live preview all appear together

Auto-Save

Once the AI finishes generating, your app is saved automatically:

• The name is generated from your prompt (e.g. "Python Fibonacci With Memoization")
• The description is your original prompt
• A confirmation banner appears at the top of the screen

Subsequent AI edits that change files are also saved automatically — you don't need to manually save after each generation round.

Fix Error

If a build error or runtime error occurs in the preview, a Fix error button appears above the chat input. Clicking it sends the error to the AI which will attempt to fix it automatically.