The PHP AI Client SDK: WordPress Calling AI vs AI Calling WordPress

Every lesson so far has covered the “AI calling WordPress” direction, abilities, MCP, the Adapter. This lesson covers the opposite direction: your own WordPress code calling out to an AI model. That’s what the PHP AI Client SDK is for, and it has nothing to do with MCP or the Abilities API, it’s a separate, standalone concern.

Step 1: The core entry point, wp_ai_client_prompt()

WordPress 7.0 ships a core function for prompting an AI model, returning a builder object you chain configuration onto:

// File: my-plugin.php
$response = wp_ai_client_prompt( 'Summarize this text in one sentence: ' . $post_content )
	->generate_text();

echo esc_html( $response );

Step 2: Which AI providers it actually supports

WordPress 7.0 ships three official provider plugins, each connecting the AI Client to one LLM vendor:

At least one provider plugin must be installed and configured with a valid API key before wp_ai_client_prompt() has anything to actually call. Which provider gets used for a given request depends on what you specify, covered next.

Step 3: Requesting a preferred model, with fallback

Rather than hardcoding one vendor, you can express a preference order, and the SDK uses the first available, configured provider from your list:

// File: my-plugin.php
$response = wp_ai_client_prompt( $prompt_text )
	->using_model_preference( 'claude-sonnet-4-5', 'gemini-3-pro-preview', 'gpt-5.1' )
	->generate_text();

This is the mechanism Course 5 uses to build provider-agnostic features, if a site owner has only configured a Gemini API key, the same code still works, it just falls through the preference list to whichever provider is actually set up.

Step 4: Embeddings, the other generation type

Beyond generating text, the SDK also supports generating embeddings, numeric vector representations of text, used for semantic search and similarity comparisons rather than for producing readable output. Embeddings live on a separate class, not on wp_ai_client_prompt():

// File: my-plugin.php
use WordPress\AiClient\AiClient;

$values = AiClient::input( $post_content )->generateEmbedding()->getValues();

Course 5 covers embeddings in full depth, and Course 10, Build a RAG Knowledge Base in WordPress, builds a complete production retrieval pipeline on top of them.

Step 5: Using the SDK outside WordPress core

If you need the same AI-calling logic in a context that isn’t a full WordPress request, a build script, a separate microservice, a CLI tool, the SDK also ships as a standalone Composer package, independent of WordPress core:

composer require wordpress/php-ai-client

use WordPress\AiClient\AiClient;

$response = AiClient::prompt( 'Summarize this text in one sentence.' )
	->generateText();

The method names differ slightly by convention (generateText() in the standalone package versus generate_text() in the WordPress-integrated core function), because the standalone package follows general PHP naming conventions rather than WordPress’s.

Test it: confirm a provider is reachable

With at least one provider plugin installed and an API key configured, run:

wp-env run cli wp eval 'echo wp_ai_client_prompt( "Say hello in five words or fewer." )->generate_text();'

You should see a short, real response from whichever provider you configured.

Recap

The PHP AI Client SDK is the “WordPress calling AI” half of the stack, unrelated to MCP or the Abilities API. wp_ai_client_prompt() is the current, core-shipped entry point (not the older AI_Client::prompt()), it supports Claude, GPT, and Gemini through official provider plugins, model preference lists for provider-agnostic code, and embeddings alongside plain text generation.

Resources & further reading

• PHP AI Client SDK repository, GitHub
• Introducing the AI Client in WordPress 7.0, make.wordpress.org
• Introducing the WordPress AI Client SDK, make.wordpress.org