The website was made by a neural network - how to publish it

What you have

The scenario is familiar: you asked ChatGPT, Claude, DeepSeek or Cursor to make a website, received a code, it even opens with a double click - but only lives on your computer. The link cannot be sent. Then the neural network usually advises “upload files to the hosting,” and that’s where the explanation ends.

Let's sort it out in order. First, determine what exactly you were given.

Option A. One HTML file

Code starts with <!doctype html>, and styles and scripts are embedded inside tags <style> and <script>. This is the most convenient case: the page is self-contained and ready to be published as is.

Option B. Multiple files

The neural network issued separately index.html, style.css and script.js. It’s also not difficult: all files should be in the same folder, and in HTML they should refer to each other using short relative paths.

Option B. Project in React, Next.js or Vue

Feature - mention npm install, file package.json, folder src. Such a project cannot simply be opened in a browser: it needs to be build. The good news is that you don’t have to build it yourself; the deployment platform will do it for you; more about this below.

How to prepare files

Three rules that solve almost all the problems of beginners.

1. Main file - index.html. Exactly, in lowercase letters, without spaces and Cyrillic. If the neural network named the file site.html or Page.html - rename.
2. All files are in one folder, index.html at its root. Not inside a subfolder: you need to load the content, not the wrapper.
3. Save in UTF-8 encoding. If you open Notepad, paste the code and save it in the default encoding, the Russian text may turn into “krakozyabry”. In Windows Notepad, the encoding is selected in the save window, in VS Code - in the lower right corner.

Check the result before publication: open index.html double click. The way the page looks locally is the same way it will appear on the Internet.

Check if there are any keys in the code

This is the only point from the article that is worth reading. Neural networks often insert into code examples blanks for access keys, and sometimes real keys that you yourself dictated to them in correspondence.

Everything that is included in the HTML or JavaScript on the page is visible to any visitor. Just click “View page code” in the browser. This does not hide any “minification” or “compression”.

Before publishing, look for the following lines in the code:

• apiKey, api_key, secret, token, password;
• long character sets starting with sk-, pk_, AKIA;
• database connection strings - these contain :// and a login:password pair.

If this is found, you need a key remove from the code and revoke in the account dashboard of the service that issued it. Consider any key that has ever been published to be compromised. Calls to paid APIs must come from the server, not from the page; Layero has it for this environment variables - they are stored in encrypted form and are not included in the page code.

How to update the site further

Edits from the neural network come in batches, so publication is not a one-time action. The update is done in the same way as the first publication: drag the folder again or run npx layero@latest deploy. The site address does not change, visitors immediately see the new version.

Each publication is saved in the deployment history. If the next edit breaks the page, the previous version is returned with one button - this is more convenient than asking the neural network to “return it as it was,” because it does not remember the exact previous version.

When the project outgrows the folder with files, it makes sense to create a repository on GitHub and link it to the project: then the site will update itself every time changes are saved, and a separate preview link will appear for each branch.