Build AI-powered applications with PHP

Model Inference

Send prompts to any AI model and get responses with a single method call. The Platform abstracts away provider differences — switch between OpenAI, Anthropic, Mistral, or local models without changing your application code.

$platform = PlatformFactory::create($apiKey, http_client());

$messages = new MessageBag(
    Message::forSystem('You are a helpful assistant.'),
    Message::ofUser('What is the Symfony framework?'),
);

$result = $platform->invoke('gpt-5-mini', $messages);
echo $result->asText();

Streaming

Stream responses token by token for real-time output. Instead of waiting for the full response, display text as it's generated — just like ChatGPT does. Works with any supported model and platform.

$result = $platform->invoke('gpt-5-mini', $messages, [
    'stream' => true,
]);

foreach ($result->asStream() as $chunk) {
    echo $chunk; // process chunks as they arrive
}

Speech

Convert text to natural-sounding speech or transcribe audio recordings to text. Build voice assistants, podcast transcription tools, or accessibility features with providers like ElevenLabs or OpenAI Whisper — all through the same interface.

// Text to Speech
$result = $platform->invoke('eleven_multilingual_v2', new Text('Hello!'), [
    'voice' => 'pqHfZKP75CvOlQylNhV4',
]);
$result->asFile('output.mp3');

// Speech to Text
$result = $platform->invoke('scribe_v1', Audio::fromFile('recording.mp3'));
echo $result->asText();

Multi-Modal

Send images, PDFs, and audio alongside text to any supported model. Build apps that can analyze photos, extract data from documents, or process voice recordings — all through the same simple API.

$messages = new MessageBag(
    Message::ofUser(
        'Describe this image',
        Image::fromFile('/path/to/photo.jpg'),
    ),
);
$result = $platform->invoke('gpt-5-mini', $messages);

echo $result->asText();

Structured Output

Map AI responses directly to PHP objects with automatic deserialization. Instead of parsing JSON strings yourself, get type-safe PHP objects back from the model — ready to use in your application logic.

$messages = new MessageBag(
    Message::forSystem('You are a helpful math tutor.'),
    Message::ofUser('Solve 8x + 7 = -23'),
);

$result = $platform->invoke('gpt-5-mini', $messages, [
    'response_format' => MathReasoning::class,
]);

$math = $result->asObject(); // MathReasoning instance

Agents

Build autonomous agents that reason, plan, and execute multi-step tasks. An Agent wraps a model with tools and processors, handling the full loop of calling tools, reading results, and deciding what to do next — until the task is complete.

$agent = new Agent($platform, 'gpt-5-mini');

$messages = new MessageBag(
    Message::forSystem('You are a helpful assistant.'),
    Message::ofUser('What is the Symfony framework?'),
);

$result = $agent->call($messages);
echo $result->getContent();

Token Usage

Track exactly how many tokens each request consumes — prompt, completion, and total. Essential for monitoring costs, optimizing prompts, and staying within budget when running AI features in production.

$result = $agent->call($messages);

$usage = $result->getMetadata()->get('token_usage');

echo $usage->getPromptTokens();
echo $usage->getCompletionTokens();
echo $usage->getTotalTokens();

Tool Calling

Let AI models call your PHP functions to fetch data or trigger actions. Define tools as simple classes — the framework handles schema generation, argument passing, and result formatting automatically.

$toolbox = new Toolbox([new YoutubeTranscriber(http_client())]);
$processor = new AgentProcessor($toolbox);
$agent = new Agent($platform, 'gpt-5-mini', [$processor], [$processor]);

$result = $agent->call($messages);

Subagents

Compose complex workflows by using agents as tools for other agents. A coordinator agent can delegate specialized tasks — like math, research, or translation — to focused subagents, each with their own model and instructions.

$mathAgent = new Agent($platform, 'gpt-5.2', [$mathPrompt]);
$subagent = new Subagent($mathAgent);

$toolbox = new Toolbox([$subagent]);
$processor = new AgentProcessor($toolbox);
$agent = new Agent($platform, 'gpt-5-mini', [$processor], [$processor]);

$result = $agent->call($messages);

RAG

Index documents into vector stores and retrieve relevant context automatically. Let your AI answer questions about your own data — manuals, blog posts, knowledge bases — by finding the most relevant passages before generating a response.

$vectorizer = new Vectorizer($platform, 'text-embedding-3-small');
$processor = new DocumentProcessor($vectorizer, $store);
$indexer = new SourceIndexer($loader, $processor);
$indexer->index(['/path/to/docs']);

$retriever = new Retriever($vectorizer, $store);
$docs = $retriever->retrieve('How do I create a command?');