Zephyr
Impreza
Guide
Support Forum

Generating Pages with AI Agents

Updated 1 day ago

Zephyr 8.46 ships with a built-in documentation set written specifically for AI assistants. Point your assistant of choice — ChatGPT, Claude, Cursor, GitHub Copilot, or any agent that can read files, and it will generate valid Zephyr page content for you in seconds, using the same shortcodes the Live Builder/WPBakery uses.

What an agent can and cannot do #

In scope — composing and editing page content from existing shortcodes and section templates. This covers the vast majority of "build me a page like X" requests.

Some shortcode attributes refer to values you configure on your own site — button styles, field styles, color schemes, and fonts. The documentation marks these as per-site values and the agent will ask you which one to use, instead of guessing.

Out of scope — the documentation set does not yet teach an agent to:

  • use WooCommerce related elements and shortcodes
  • edit Display Logic settings
  • edit Scrolling Effects settings
  • edit Theme Options (global colors, sizes and typography)
  • edit a Header (site top area)
  • edit a Grid Layout
  • create customizations

What is included #

After you install or update Zephyr, a new folder appears inside the plugin:

wp-content/plugins/us-core/llms/

It is a self-contained, AI-readable reference of everything an assistant needs to compose a page. The files are plain Markdown and plain text. You can open them in any editor; the agent reads them the same way you would.

Without this documentation, an AI assistant has to guess at shortcode names, parameters, and nesting rules. The result is usually markup that looks like Impreza but won't open in the builder, won't render responsively, or silently falls back to defaults.

With llms/ in front of it, the assistant works from the same parameter definitions the plugin itself uses. The output is markup you can paste straight into a page and edit visually afterwards.

How to use it with your AI assistant #

There are three common setups. Pick the one that matches the tool you already use.

Option 1 — ChatGPT, Claude, or Gemini (web chat) #

  1. On your server, locate the folder wp-content/plugins/us-core/llms/.
  2. Download it to your computer (via FTP, file manager, or the hosting control panel).
  3. Drag and drop the contents into the chat window as attachments. For a typical page you only need llms.txt, composition-rules.md, and shortcodes.md. Add a file from sections/ if you want to start from a pre-built section.
  4. Ask the assistant for the page you want — see the example prompt below.
  5. Copy the generated [vc_row]…[/vc_row] markup and paste it into a new page.

Option 2 — Cursor, Claude Code, or another IDE agent #

  1. Open the folder that contains your WordPress installation (or just the us-core plugin) in your IDE.
  2. The agent will index plugins/us-core/llms/ automatically. Some tools also pick up llms.txt as a project context file.
  3. In the chat panel, ask for the page. The agent will read the relevant documentation files on demand.
  4. Copy the generated[vc_row]…[/vc_row] markup and paste it into a new page.

Option 3 — Direct via the WordPress REST API #

For agencies and developers running automated workflows, an AI agent can publish the generated content directly using the standard WordPress REST endpoint /wp-json/wp/v2/pages with an Application Password. The shortcode markup goes into the content field. No additional plugin or MCP server is required.

Quick start: ask for a page #

Once your assistant has access to the llms/ folder, a prompt like the one below is enough to produce a full page:

Using the documentation in the us-core llms folder, generate the
post_content for a landing page with these sections:

1. Hero with a heading, a one-line subtitle, and two buttons
   (primary "Get started", secondary "See pricing")
2. Three feature columns with icons
3. A pricing table with three plans
4. An FAQ section with five questions
5. A final call-to-action banner

Use the Impreza palette tokens for colors. Output only the raw
shortcode markup, no explanation.

The assistant returns a block of [vc_row]…[/vc_row] markup. Create a new page in the WordPress admin, switch the editor to Backend Editor or paste into the Text tab of the Classic Editor, drop the markup in, save, and open the page in the live builder to fine-tune visually.

FAQ #

Do I need a special plugin or API key? #

No. The llms/ folder ships inside us-core starting with version 8.46. Any AI assistant that can read files can use it.

Will the generated markup work in the Live Builder / WPBakery Builder? #

Yes. The documentation is generated from the same element configurations the builder uses, so a page produced by an agent opens and edits in the builder like any other page.

The agent produced markup that looks broken. What now? #

Make sure the assistant actually has access to composition-rules.md and shortcodes.md — these two files contain the rules most often missed when an agent is guessing. If you are working in a web chat, re-attach them and ask again.

In other cases please create a support ticket and attach the generated content.

Will this folder be updated? #

Yes. Each file declares the UpSolution Core version it was generated from. The documentation is regenerated on each plugin release, so updating the UpSolution Core updates the AI-readable docs at the same time.

Can I delete the llms/ folder to save space? #

You can, but there is no sense. The folder is small (under 2 MB) and has no effect on site performance — it is never loaded by WordPress at runtime.