Skip to main content
Inference is a request facade over InferenceRuntime. Use it to build requests fluently, then execute as text, response object, JSON, or stream.

​
Create an Instance

<?php
use Cognesy\Polyglot\Inference\Inference;

$default = new Inference();
$openai = Inference::using('openai');
$fromDsn = Inference::fromDsn('preset=openai,model=gpt-4.1-nano');
// @doctest id="3c39"

​
Build a Request

<?php
use Cognesy\Polyglot\Inference\Inference;

$inference = (new Inference())
    ->withMessages('Summarize CQRS in two sentences.')
    ->withModel('gpt-4.1-nano')
    ->withOptions(['temperature' => 0.2])
    ->withMaxTokens(120)
    ->withStreaming(false);
// @doctest id="1593"

​
Execute

<?php
$text = $inference->get();          // string
$response = $inference->response(); // InferenceResponse
$json = $inference->asJson();       // JSON string
$data = $inference->asJsonData();   // array
$stream = $inference->stream();     // InferenceStream
// @doctest id="eae9"

​
One-Call Configuration

<?php
use Cognesy\Polyglot\Inference\Enums\OutputMode;

$text = (new Inference())
    ->with(
        messages: 'Return valid JSON with key "status".',
        options: ['temperature' => 0],
        mode: OutputMode::Json,
    )
    ->get();
// @doctest id="1cda"

​
Runtime-First Usage

<?php
use Cognesy\Polyglot\Inference\Data\InferenceRequest;
use Cognesy\Polyglot\Inference\Inference;

$runtime = Inference::using('openai')->runtime();

$pending = $runtime->create(new InferenceRequest(messages: 'Ping', model: 'gpt-4.1-nano'));
$text = $pending->get();
// @doctest id="44ab"

​
Register Custom Drivers

<?php
use Cognesy\Http\HttpClient;
use Cognesy\Polyglot\Inference\Config\LLMConfig;
use Cognesy\Polyglot\Inference\Contracts\CanProcessInferenceRequest;
use Cognesy\Polyglot\Inference\Drivers\OpenAI\OpenAIDriver;
use Cognesy\Polyglot\Inference\Inference;
use Psr\EventDispatcher\EventDispatcherInterface;

Inference::registerDriver(
    'openai-custom',
    function (
        LLMConfig $config,
        HttpClient $httpClient,
        EventDispatcherInterface $events,
    ): CanProcessInferenceRequest {
        return new OpenAIDriver($config, $httpClient, $events);
    }
);

Inference::unregisterDriver('openai-custom');
Inference::resetDrivers();
// @doctest id="42bd"

​
See Also