README

Official Mistral provider for the PHP AI SDK. Uses Mistral's OpenAI-compatible Chat Completions API and embeddings endpoint.

Installation

composer require aisdk/mistral

Basic Usage

use AiSdk\Generate;
use AiSdk\Mistral;

$result = Generate::text()
    ->model(Mistral::model('mistral-large-latest'))
    ->instructions('Write short, clear answers.')
    ->prompt('Explain closures in PHP.')
    ->run();

echo $result->text;

Default model shorthand:

Generate::model(Mistral::model('mistral-large-latest'));

$result = Generate::text('Explain closures in PHP.')->run();

Configuration

Environment Variables

Programmatic Configuration

$provider = Mistral::create([
    'apiKey' => 'mistral-...',
    'baseUrl' => 'https://api.mistral.ai/v1',
    'headers' => ['X-Custom-Header' => 'value'],
]);

Supported Capabilities

Streaming

use AiSdk\Generate;
use AiSdk\Mistral;

$stream = Generate::text('Tell me a story.')
    ->model(Mistral::model('mistral-large-latest'))
    ->stream();

foreach ($stream->chunks() as $chunk) {
    echo $chunk;
}

$result = $stream->run();

Embeddings

use AiSdk\Generate;
use AiSdk\Mistral;

$result = Generate::embedding(['Search query', 'Document text'])
    ->model(Mistral::model('mistral-embed'))
    ->providerOptions('mistral', ['user' => 'user-123'])
    ->run();

$queryVector = $result->embeddings[0]->vector;
$documentVector = $result->embeddings[1]->vector;

Mistral's embedding request schema does not expose a dimensions field, so this adapter rejects the portable dimensions() option before sending a request.

Structured Output

Without an exact runtime capability override, the provider adapter degrades json_schema to json_object with an injected JSON instruction:

use AiSdk\Generate;
use AiSdk\Mistral;
use AiSdk\Schema;

$result = Generate::text()
    ->model(Mistral::model('mistral-large-latest'))
    ->prompt('Extract the city and country from: Lahore, Pakistan.')
    ->output(Schema::object(
        name: 'address',
        properties: [
            Schema::string(name: 'city')->required(),
            Schema::string(name: 'country')->required(),
        ],
    ))
    ->run();

Models with native json_schema support (openai/gpt-oss-20b, openai/gpt-oss-120b, moonshotai/kimi*) use the native format directly.

Tools

use AiSdk\Generate;
use AiSdk\Mistral;
use AiSdk\Schema;
use AiSdk\Tool;

$weather = Tool::make('weather', 'Get current weather')
    ->input(Schema::string(name: 'city')->required())
    ->run(fn (string $city): string => "Sunny in {$city}");

$result = Generate::text()
    ->model(Mistral::model('mistral-large-latest'))
    ->prompt('What is the weather in Lahore?')
    ->tool($weather)
    ->run();

Model IDs and Capabilities

Mistral model IDs pass through unchanged and do not need to be registered. The package does not ship a model inventory; the Mistral API remains the authority on whether a particular model accepts a requested feature.

Capabilities describe what the Mistral adapter can serialize. The Mistral API returns a normalized SDK exception if the selected model or requested feature is rejected.

Provider-Specific Options

Raw provider options can be passed as an escape hatch:

$result = Generate::text('Hello')
    ->model(Mistral::model('mistral-large-latest'))
    ->providerOptions('mistral', [
        'raw' => ['top_k' => 40],
    ])
    ->run();

Testing

composer test

Links

• Core Package
• OpenAI-Compatible Package
• Mistral API documentation
• Mistral Python SDK Embeddings Resource
• Project Documentation