Prism Workers AI — A Cloudflare Workers AI Provider for Prism PHP

Prism Workers AI is a Cloudflare Workers AI provider for Prism PHP and the Laravel AI SDK. It routes requests through Cloudflare's AI Gateway /compat endpoint.

• Generate natural language text, structured outputs, and stream responses.
• Create vector representations using Cloudflare embedding models.
• Automate tool execution with support for complex, multi-step workflows.
• Extract and utilize reasoning content from supported reasoning models.
• Maintain conversation context using session affinity for prompt prefix caching.
• Compatible with Prism PHP and Laravel AI SDK's agent() helper.

#Text Generation and Streaming

Generate text using any Workers AI model through Prism's fluent API:

use Prism\Prism\Facades\Prism;
 
$response = Prism::text()
    ->using('workers-ai', 'workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast')
    ->withPrompt('Hello!')
    ->asText();

Streaming works the same way:

$stream = Prism::text()
    ->using('workers-ai', 'workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast')
    ->withPrompt('Tell me a story')
    ->asStream();

#Embeddings

Unlike Prism's built-in xAI driver, this package supports embeddings:

$response = Prism::embeddings()
    ->using('workers-ai', 'workers-ai/@cf/baai/bge-large-en-v1.5')
    ->fromInput('Hello world')
    ->generate();

#Tool Calling

The package supports tool calling with multi-step workflows. Workers AI requires content to always be present on assistant messages, even when empty during tool call responses — this package handles that automatically:

$response = Prism::text()
    ->using('workers-ai', 'workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast')
    ->withTools([$weatherTool])
    ->withMaxSteps(3)
    ->withPrompt('What is the weather?')
    ->asText();

#Reasoning Models with Thinking Content

The package extracts reasoning content from reasoning models such as Kimi K2.5. The reasoning chain is available alongside the final response:

$response = Prism::text()
    ->using('workers-ai', 'workers-ai/@cf/moonshotai/kimi-k2.5')
    ->withMaxTokens(2000)
    ->withPrompt('What is 15 * 37?')
    ->asText();
 
$response->text; // "555"
$response->steps[0]->additionalContent['thinking'];
// "The user is asking for the product of 15 and 37..."

When streaming, you receive thinking events before the text output:

$stream = Prism::text()
    ->using('workers-ai', 'workers-ai/@cf/moonshotai/kimi-k2.5')
    ->withMaxTokens(2000)
    ->withPrompt('Explain briefly why the sky is blue.')
    ->asStream();
 
foreach ($stream as $event) {
    // ThinkingStartEvent, ThinkingEvent (deltas), ThinkingCompleteEvent
    // then TextStartEvent, TextDeltaEvent, TextCompleteEvent
}

Set withMaxTokens(2000) or higher when using reasoning models, since reasoning tokens count against the limit.

#Laravel AI SDK Integration

If you're using laravel/ai, the package also registers a workers-ai driver:

use function Laravel\Ai\agent;
 
$response = agent(instructions: 'You are a helpful assistant.')
    ->prompt('Hello!', provider: 'workers-ai');

You can learn more about this package on GitHub at meirdick/prism-workers-ai.

Note: the package requires PHP 8.2+ and Prism PHP ^0.99, with optional support for Laravel AI SDK ^0.3.