# Getting started (https://lazyfrog.dev/docs)

LazyFrog is where the apps your AI builds live. You talk to your own AI, such as Claude or ChatGPT, and LazyFrog gives it a private workspace, a database, a live preview and a way to publish. You never need a terminal.

## 1. Connect your AI [#1-connect-your-ai]

Add LazyFrog to your AI once as a connector. [Connect your AI](/connect) has step-by-step instructions for Claude, Claude Desktop, Claude Code, ChatGPT and Codex. You'll need a LazyFrog account: [start free](https://app.lazyfrog.dev/sign-up), then sign in when your AI asks.

## 2. Ask for something [#2-ask-for-something]

Start a new chat and say what you want, as you would to a person:

> Make a website for my bakery in York, with the menu, opening hours and ordering for tomorrow.

A good AI asks a few questions first (what you sell, the look you like) and proposes a short plan. Then it creates a project and builds it.

## 3. Watch it in the preview [#3-watch-it-in-the-preview]

Every project has a **preview address** like `https://p-7k2q9x.lazyfrog.app`. Your AI shares it with you. Open it on your phone or laptop and it updates as the AI works. Ask for changes in plain words: "make the header green", "add a page for catering".

## 4. Publish [#4-publish]

When you like it, say "publish it" or press **Publish** in the [portal](https://app.lazyfrog.dev). It goes live at `https://<name>.lazyfrog.app` with no downtime, and every later publish switches over the same way, on [your own domain](/docs/domains) too if you connect one. Publishing needs a paid plan: Free is for building and trying, and its preview is the one address you share.

## What you see in the portal [#what-you-see-in-the-portal]

The portal at [app.lazyfrog.dev](https://app.lazyfrog.dev) shows each project: what your AI did, its preview and live site, the database, files, snapshots and deployments, and buttons to publish, undo and export. You can do everything there that your AI can, apart from writing code.

## If something goes wrong [#if-something-goes-wrong]

Every risky step takes a [snapshot](/docs/snapshots) first, so you can always go back. Ask your AI to undo, or restore a snapshot in the portal. If the platform itself is at fault, your AI can open a support request for you.