WordPress AI Client

Part of the AI Building Blocks for WordPress initiative

An AI client and API for WordPress to communicate with any generative AI models of various capabilities using a uniform API.

Built on top of the PHP AI Client, adapted for the WordPress ecosystem.

Features

• WordPress-native Prompt Builder: Fluent API for building and configuring AI prompts, built directly on top of the PHP AI Client while following WordPress Coding Standards and best practices.
• Admin Settings Screen: Integrated settings screen in WP Admin to provision AI provider API credentials.
• Automatic Credential Wiring: Automatic wiring up of AI provider API credentials based on storage in a WordPress database option.
• PSR-compliant HTTP Client: HTTP client implementation using the WordPress HTTP API, fully compatible with PSR standards.

Installation

composer require wordpress/wp-ai-client

Configuration

1. Initialize the Client

You must initialize the client on the WordPress init hook. This sets up the HTTP client integration and registers the settings screen.

add_action( 'init', array( 'WordPress\AI_Client\AI_Client', 'init' ) );

2. Configure API Credentials

Before making requests, you need to configure API keys for your desired providers (e.g. Anthropic, Google, OpenAI).

1. Go to Settings > AI Credentials in the WordPress Admin.
2. Enter your API keys for the providers you intend to use.
3. Save changes.

Usage

The SDK provides a fluent Prompt_Builder interface to construct and execute AI requests.

Text Generation

use WordPress\AI_Client\AI_Client;

$text = AI_Client::prompt( 'Write a haiku about WordPress.' )
	->generate_text();

echo wp_kses_post( $text );

Image Generation

use WordPress\AI_Client\AI_Client;

$image_file = AI_Client::prompt( 'A futuristic WordPress logo in neon style' )
	->generate_image();

$data_uri = $image_file->getDataUri();

echo '<img src="' . esc_url( $data_uri ) . '" alt="A futuristic WordPress logo in neon style">';

Advanced Usage

JSON Output and Temperature

use WordPress\AI_Client\AI_Client;

$schema = array(
	'type'       => 'array',
	'items'      => array(
		'type'       => 'object',
		'properties' => array(
			'plugin_name' => array( 'type' => 'string' ),
			'category'    => array( 'type' => 'string' ),
		),
		'required'   => array( 'plugin_name', 'category' ),
	),
);

$json = AI_Client::prompt( 'List 5 popular WordPress plugins with their primary category.' )
	->using_temperature( 0.2 ) // Lower temperature for more deterministic result.
	->as_json_response( $schema )
	->generate_text();

// Output will be a JSON string adhering to the schema.
$data = json_decode( $json, true );

Generating Multiple Image Candidates

use WordPress\AI_Client\AI_Client;

$images = AI_Client::prompt( 'Aerial shot of snowy plains, cinematic.' )
	->generate_images( 4 );

foreach ( $images as $image_file ) {
	echo '<img src="' . esc_url( $image_file->getDataUri() ) . '" alt="Aerial shot of snowy plains">';
}

Multimodal Output (Text & Image)

use WordPress\AI_Client\AI_Client;
use WordPress\AiClient\Messages\Enums\ModalityEnum;

$result = AI_Client::prompt( 'Create a recipe for a chocolate cake and include photos for the steps.' )
	->as_output_modalities( ModalityEnum::text(), ModalityEnum::image() )
	->generate_result();

// Iterate through the message parts.
foreach ( $result->toMessage()->getParts() as $part ) {
	if ( $part->isText() ) {
		echo wp_kses_post( $part->getText() );
	} elseif ( $part->isFile() && $part->getFile()->isImage() ) {
		echo '<img src="' . esc_url( $part->getFile()->getDataUri() ) . '" alt="">';
	}
}

Best Practices

Automatic Model Selection

By default, the SDK automatically chooses a suitable model based on the prompt's requirements (e.g., text vs. image) and the configured providers on the site. This makes your plugin provider-agnostic, allowing it to work on any site regardless of which AI provider the admin has configured.

Using Model Preferences

If you prefer specific models for better performance or capabilities, use using_model_preference(). The SDK will try to use the first available model from your list. If none are available (e.g., provider not configured), it falls back to automatic selection.

Pass preferences as an array of [ provider_id, model_id ] to ensure the correct provider is targeted.

use WordPress\AI_Client\AI_Client;

$summary = AI_Client::prompt( 'Summarize the history of the printing press.' )
	->using_temperature( 0.1 )
	->using_model_preference(
		array( 'anthropic', 'claude-sonnet-4-5' ),
		array( 'google', 'gemini-3-pro-preview' ),
		array( 'openai', 'gpt-5.1' )
	)
	->generate_text();

Using a Specific Model

Enforcing a single specific model using using_model() restricts your feature to sites that have that specific provider configured. For most scenarios, this is unnecessarily opinionated. Only use this approach if you really only want to offer the feature in combination with that model.

use WordPress\AI_Client\AI_Client;
use WordPress\AiClient\ProviderImplementations\Anthropic\AnthropicProvider as Anthropic;

$text = AI_Client::prompt( 'Explain quantum computing in simple terms.' )
	->using_model( Anthropic::model( 'claude-sonnet-4-5' ) )
	->generate_text();

Feature Detection

Before actually sending an AI prompt and getting a response, always check if the prompt is supported before execution.

This is always recommended, but especially crucial if you require the use of a specific model.

use WordPress\AI_Client\AI_Client;

$prompt = AI_Client::prompt( 'Explain quantum computing in simple terms.' )
	->using_temperature( 0.2 );

if ( $prompt->is_supported_for_text_generation() ) {
	// Safe to generate.
	$text = $prompt->generate_text();
} else {
	// Fallback: Hide feature or show setup instructions.
}

The above condition will only evaluate to true if the site has one or more providers configured with models that support text generation including a temperature configuration.

Generally, using is_supported_for_text_generation() (or is_supported_for_image_generation(), etc.) ensures you only expose AI features that can actually run on the current site configuration.

Error Handling

The SDK offers two ways to handle errors.

1. Exception Based

AI_Client::prompt() throws exceptions on failure.

try {
	$text = AI_Client::prompt( 'Hello' )->generate_text();
} catch ( \Exception $e ) {
	wp_die( $e->getMessage() );
}

echo wp_kses_post( $text );

2. WP_Error Based

AI_Client::prompt_with_wp_error() returns a WP_Error object on failure.

Note: This error handling mechanism is experimental. We are gathering feedback to decide whether the SDK should primarily focus on exceptions or WP_Error objects.

$text = AI_Client::prompt_with_wp_error( 'Hello' )
	->generate_text();

if ( is_wp_error( $text ) ) {
	wp_die( $text->get_error_message() );
}

echo wp_kses_post( $text );

Architecture

This library is a WordPress-specific wrapper around the PHP AI Client.

• WordPress\AI_Client\AI_Client: The main entry point.
• WordPress\AI_Client\Builders\Prompt_Builder: A fluent builder for constructing AI requests. It maps WordPress-style snake_case methods to the underlying SDK's camelCase methods.
• WordPress\AI_Client\Builders\Prompt_Builder_With_WP_Error: A wrapper around Prompt_Builder that catches exceptions and returns WP_Error objects.

Further reading

See the Prompt_Builder class and its public methods for all the ways you can configure the prompt.

See the contributing documentation for more information on how to get involved.