> ## Documentation Index
> Fetch the complete documentation index at: https://docs.instructorphp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Openai compatible reasoning content

## Overview

The `openai-compatible` driver extracts reasoning returned in the assistant
message's `reasoning_content` field. This also works for an application model
that has no bundled catalog record because reading response content does not
require a typed reasoning selection.

Set `OPENAI_COMPATIBLE_API_URL`, `OPENAI_COMPATIBLE_API_KEY`, and
`OPENAI_COMPATIBLE_MODEL` for your backend. The live example otherwise uses
DeepSeek through its OpenAI-compatible endpoint.

## Example

```php theme={null}
<?php
require 'examples/boot.php';

use Cognesy\Config\Env;
use Cognesy\Messages\Messages;
use Cognesy\Polyglot\Inference\Config\LLMConfig;
use Cognesy\Polyglot\Inference\Inference;
use Cognesy\Utils\Str;

$customApiUrl = (string) Env::get('OPENAI_COMPATIBLE_API_URL', '');
$usesCustomEndpoint = $customApiUrl !== '';
$apiUrl = $usesCustomEndpoint ? $customApiUrl : 'https://api.deepseek.com';
$apiKey = (string) Env::get(
    'OPENAI_COMPATIBLE_API_KEY',
    $usesCustomEndpoint ? 'local' : Env::get('DEEPSEEK_API_KEY', ''),
);
$model = (string) Env::get(
    'OPENAI_COMPATIBLE_MODEL',
    $usesCustomEndpoint ? 'local-reasoner' : 'deepseek-v4-flash',
);

$config = new LLMConfig(
    apiUrl: $apiUrl,
    apiKey: $apiKey,
    endpoint: '/chat/completions',
    model: $model,
    maxTokens: 128,
    driver: 'openai-compatible',
);

$response = Inference::fromConfig($config)
    ->withMessages(Messages::fromString(
        'What is the capital of France? Answer with just the city name.',
    ))
    ->response();

$message = $response->message();

echo "ASSISTANT: {$message->content()->toString()}\n";
echo "REASONING: {$message->reasoningContent()}\n";

assert(Str::contains($message->content()->toString(), 'Paris'));
assert($message->reasoningContent() !== '');
?>
```