Basics

¶Chat API Documentation

¶Introduction

php-mistral provides a unified interface for sending messages to multiple LLM providers. Originally designed for Mistral Platform, the library ensures compatibility across various services while maintaining simplicity.

¶Example with Mistral Platform

use Partitech\PhpMistral\Clients\Mistral\MistralClient;

$apiKey   = getenv('MISTRAL_API_KEY');
$client   = new MistralClient($apiKey);
$messages = $client ->getMessages()
                    ->addSystemMessage(content: 'You are a gentle bot who respond like a pirate')
                    ->addUserMessage(content: 'What is the best French cheese?');

$params = [
    'model' => 'mistral-large-latest',
    'temperature' => 0.7,
    'top_p' => 1,
    'max_tokens' => 512,
];

// Non-streaming
$response = $client->chat(messages: $messages, params: $params);
echo $response->getMessage();

// Streaming
foreach ($client->chat(messages: $messages, params: $params, stream: true) as $chunk) {
    echo $chunk->getChunk();
}

¶Client Instantiation

Here’s how to instantiate the client for each supported provider:

use Partitech\PhpMistral\Clients\Mistral\MistralClient;

$client = new MistralClient(apiKey: getenv('MISTRAL_API_KEY'));

use Partitech\PhpMistral\Clients\Anthropic\AnthropicClient;

$client = new AnthropicClient(apiKey: getenv('ANTHROPIC_API_KEY'));

  use Partitech\PhpMistral\Clients\HuggingFace\HuggingFaceClient;

  $client = new HuggingFaceClient(
      apiKey: (string) $apiKey, 
      provider: 'hf-inference', 
      useCache: true, 
      waitForModel: true
  );

  use Partitech\PhpMistral\Clients\Tgi\TgiClient;

  $client = new TgiClient(apiKey: (string) $apiKey, url: $tgiUrl);

use Partitech\PhpMistral\Clients\LlamaCpp\LlamaCppClient;

$client = new LlamaCppClient(
      apiKey: $llamacppApiKey, 
      url: $llamacppUrl
);

use Partitech\PhpMistral\Clients\Ollama\OllamaClient;

$client = new OllamaClient(url: getenv('OLLAMA_URL'));

use Partitech\PhpMistral\Clients\Vllm\VllmClient;

$client = new VllmClient(
  apiKey: getenv('VLLM_API_KEY'), 
  url: getenv('VLLM_URL')
);

use Partitech\PhpMistral\Clients\Xai\XaiClient;

$client = new XaiClient(apiKey: getenv('XAI_API_KEY'));

¶Usage of chat() Method

Once your client is instantiated, the usage of chat() is identical across all providers.

¶Non-Streaming Example

$messages = $client->getMessages() 
                   ->addSystemMessage(content: 'You are a gentle bot who respond like a pirate')
                   ->addUserMessage(content: 'Tell me a fun fact about cheese.');

$params = [
    'model' => 'your-model-name',
    'temperature' => 0.7,
    'top_p' => 1,
    'max_tokens' => 512,
];

$response = $client->chat(messages: $messages, params: $params);
echo $response->getMessage();

¶Streaming Example

$messages = $client->getMessages() 
                   ->addSystemMessage(content: 'You are a gentle bot who respond like a pirate')
                   ->addUserMessage(content: 'Tell me a fun fact about cheese.');

$params = [
    'model' => 'your-model-name',
    'temperature' => 0.7,
    'top_p' => 1,
    'max_tokens' => 512,
];

foreach ($client->chat(messages: $messages, params: $params, stream: true) as $chunk) {
    echo $chunk->getChunk();
}

Note

Replace 'your-model-name' with the appropriate model for your provider.
Refer to the parameters table for provider-specific options.

¶Parameters comparison

¶Common Parameters

These parameters are supported by all providers and should be prioritized for cross-provider compatibility:

¶Mistral Platform Parameters

¶Anthropic Parameters

¶Hugging Face / TGI Parameters

¶Llama.cpp Parameters

¶Ollama Parameters

¶vLLM Parameters

¶XAI Parameters