README

Overview

A PHP library that provides a unified interface for interacting with multiple AI services (OpenAI, Anthropic, Mistral, Groq, xAI). Simplifies integration and standardizes interactions across different AI providers while maintaining service-specific features.

Implementations

Requirements

• PHP 8.2+
• Composer

Installation

composer require codechap/ai

Basic Usage

OpenAI

use codechap\ai\Ai;
$openai = new ai('openai', $openaiKey);
print $openai
    ->set('temperature', 0)
    ->set('model', 'o3-mini-2025-01-31')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('json', true)
    ->set('reasoningEffort', 'low')
    ->query("What is the capital of South Africa? Only return the three in a JSON response.")
    ->one()
    ;

Mistral

use codechap\ai\Ai;
$mistral = new Ai('mistral', $mistralKey);
print $mistral
    ->set('temperature', 0)
    ->set('model', 'o3-mini-2025-01-31')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('json', true)
    ->query("What is the capital of South Africa? Only return the three in a JSON response.")
    ->one()
    ;

Groq (Most open source models)

use codechap\ai\Ai;
$groq = new Ai('groq', $groqKey);
print $groq
    ->set('temperature', 0)
    ->set('model', 'deepseek-r1-distill-llama-70b')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('stream', false)
    ->query("What is the capital of South Africa?")
    ->one()
    ;
print "\n\n";

Anthropic (Claude)

use codechap\ai\Ai;
$anthropic = new Ai('anthropic', $anthropicKey);
print $anthropic
    ->set('temperature', 0)
    ->set('model', 'claude-3-5-sonnet-20241022')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('stream', false)
    ->query("What is the capital of South Africa?")
    ->one()
    ;

xAI (Grok)

use codechap\ai\Ai;
$xai = new Ai('xai', $xaiKey);
print $xai
    ->set('temperature', 0)
    ->set('model', 'grok-2-latest')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('stream', false)
    ->query("What is the capital of South Africa?")
    ->one()
    ;

Google (Gemini)

use codechap\ai\Ai;
$google = new Ai('google', $googleKey);
print = $google
    ->set('temperature', 0)
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('stream', false)
    ->set('json', true)
    ->query("What is the capital of South Africa? Only return the three in a JSON response.")
    ->all()
    ;

Vision Example

$groq = new Ai('groq', $groqKey);
$result = $groq
    ->set('temperature', 0)
    ->set('model', 'meta-llama/llama-4-scout-17b-16e-instruct')
    ->set('systemPrompt', 'You are a helpful assistant from planet earth.')
    ->set('stream', false)
    ->set('json', false)
    ->query(
    [
        [
            'role' => 'user',
            'content' => [
                [
                    'type' => 'image_url',
                    'image_url' => [
                        'url' => 'https://upload.wikimedia.org/wikipedia/commons/f/f2/LPU-v1-die.jpg'
                    ]
                ],
                [
                    'type' => 'text',
                    'text' => 'What is this image about?'
                ]
            ]
        ]
    ])
    ->all()
    ;
print_r($result);

JSON Response Handling

Different AI services handle JSON responses in different ways:

OpenAI

• Uses native JSON response formatting via the response_format parameter
• Set json: true to automatically receive properly formatted JSON responses
• No additional processing needed

Other Services (Anthropic, Mistral, Groq, xAI)

• JSON responses are handled through post-processing
• Set json: true to enable JSON extraction and validation
• Uses the JsonExtractor helper to:
  • Extract JSON from raw responses
  • Handle JSON within markdown code blocks (json ... )
  • Validate JSON structure

Example usage:

// OpenAI (native JSON)
$ai->openai()
   ->set('json', true)
   ->query('Return user data')
   ->one();

// Other services (post-processed JSON)
$ai->anthropic() // or mistral(), groq(), xai()
   ->set('json', true)
   ->query('Return user data')
   ->one();

Scope

This library is focused on text-based chat completions across multiple AI providers.

Image generation is intentionally out of scope and will not be added here. It is better suited as a separate project (e.g., an MCP server) due to its fundamentally different request/response patterns.

Contributing

• Todo

Testing

• PHPUnit test suite
• Automatic service discovery testing
• Error handling verification

Contributing by adding a New Service

To add a new AI service:

1. Create a new file in src/Services/ following the naming convention

2. Implement the required methods:

   • __construct(string $apiKey)
   • query(string $prompt): string

3. The service will be automatically discovered and available through AIWrapper

4. Run composer test to verify your implementation