Zactonz AI Connector for xAI Grok

Adds an xAI Grok connector to Settings > Connectors for the WordPress AI Client, including image generation.

xAI publishes the Grok family for text, vision, and image generation. This connector reads the richer xAI model endpoint, so each Grok model arrives in WordPress with the input and output modalities xAI reports, and image models are wired to the image generation endpoint rather than chat.

WordPress 7.0 introduced the AI Client and a Settings > Connectors screen where every provider's credentials live in one place. Core ships OpenAI, Anthropic and Google. This plugin adds xAI Grok to that same screen, so anything built on wp_ai_client_prompt() can use it without knowing anything about this plugin.

Version 1.0.0 · Developer: Zactonz Technologies

Documentation: developers.zactonz.com · Releases · Report an issue

xAI Grok is a third-party service. This plugin is developed by Zactonz Technologies and is not affiliated with, endorsed by or sponsored by its owner. xAI and Grok are trademarks of X.AI Corp.

Requirements

• WordPress 7.0 or higher
• PHP 7.4 or higher
• An xAI Grok account and API key

Installation

1. Copy this directory to wp-content/plugins/zactonz-ai-connector-for-xai-grok/.
2. Activate Zactonz AI Connector for xAI Grok from the Plugins screen.
3. Add the credentials on Settings > Connectors, or on Settings > xAI Grok.

Credentials

You can supply the API key in three ways. They are checked in this order, and the first one found wins:

1. A PHP constant, defined before WordPress loads — normally in wp-config.php:

   define( 'XAI_API_KEY', 'your-api-key' );

2. An environment variable of the same name:

   export XAI_API_KEY="your-api-key"

3. The admin field on Settings > Connectors, stored in its own option and never rendered back into the page.

When a constant or environment variable is set, the admin field is disabled and labelled as overridden, so nobody edits a value that cannot take effect. This is the recommended setup for version-controlled or multi-environment sites: the key stays out of the database and out of any export of it.

Get a key at console.x.ai.

Usage

Once the connector is configured, use it through the WordPress AI Client by provider ID. Nothing in your code needs to reference this plugin:

$text = wp_ai_client_prompt( 'Write a short WordPress release note.' )
    ->using_provider( 'xai' )
    ->generate_text();

To pin a specific model rather than the configured default:

use Zactonz\AiConnectorForXAi\Provider\XAiProvider;

$model  = XAiProvider::model( 'your-model-id' );
$result = $model->generateTextResult(
    array( new UserMessage( array( new MessagePart( 'Write a release note.' ) ) ) )
);

Streaming

The AI Client has no provider-neutral streaming contract yet, so the text model exposes generateStreamResult(). Each event has a type of content_delta, thinking_delta, tool_call_delta or done. Return false from the callback to stop and receive the partial result.

use WordPress\AiClient\Messages\DTO\MessagePart;
use WordPress\AiClient\Messages\DTO\UserMessage;
use Zactonz\AiConnectorForXAi\Provider\XAiProvider;

$model  = XAiProvider::model( 'your-model-id' );
$result = $model->generateStreamResult(
    array( new UserMessage( array( new MessagePart( 'Write a release note.' ) ) ) ),
    static function ( array $event ) {
        if ( 'content_delta' === $event['type'] ) {
            echo esc_html( $event['delta'] );
            flush();
        }
    }
);

Streaming is an ordinary wp_remote_post(). The body arrives through the requests-request.progress action that WordPress has bridged from the Requests library since 4.7, so there is no cURL call anywhere in this plugin. Where a host has no transport that delivers incrementally, the whole response is replayed through the same parser: every event still arrives, just at once.

Configuration

Model discovery

Models are listed from the xAI API.

Capabilities are read from the provider's own model listing rather than guessed. A model is advertised as vision, tool-calling, structured-output or embedding capable only when the provider reports it, which keeps the WordPress model picker honest.

Diagnostics

Settings > xAI Grok has a diagnostics button reporting the endpoint, HTTP status, latency, number of discovered models, any configured defaults missing from the current model list, and the AI Client version. Credentials never appear in the report. The same check is registered as a WordPress Site Health test.

Development

composer install
composer lint

composer lint runs PHP_CodeSniffer against the WordPress standards, PHPStan at level 6, and the PHPUnit suite.

Changelog

See CHANGELOG.md.

Security

See SECURITY.md for how to report a vulnerability.

License

GPL-2.0-or-later. See LICENSE.