cheinisch/openrouter-php-client

Modern, lightweight PHP client for the OpenRouter.ai API. Provides a simple interface for chat completions, text generation, and custom LLM endpoints.

v0.3 2026-04-21 18:33 UTC

Requires

• php: >=8.1
• guzzlehttp/guzzle: ^7.0

Requires (Dev)

• phpunit/phpunit: ^10.0

MIT

• php
• client
• ai
• llm
• OpenRouter
• chat-completions

README

A lightweight PHP client for the OpenRouter.ai API.
Supports simple single-line calls as well as advanced features like tool calling, structured outputs, SSE streaming, plugins, provider routing, and reasoning models.

Requirements

• PHP >= 8.1
• Composer
• guzzlehttp/guzzle (installed automatically)

Installation

composer require cheinisch/openrouter-php-client

Usage

1) Static convenience method (minimal)

<?php
require __DIR__.'/vendor/autoload.php';
use cheinisch\OpenRouterClient;

$apiKey = getenv('OPENROUTER_API_KEY');

echo OpenRouterClient::OpenRouterChat($apiKey, 'openai/gpt-4o-mini', 'Say only: OK');

2) With optional attribution headers

echo OpenRouterClient::OpenRouterChat(
    $apiKey,
    'mistralai/mistral-small',
    'Give me one short fun fact about PHP.',
    'https://my-app.com',  // optional – HTTP-Referer
    'My App'               // optional – X-Title
);

3) Client instance – simple chat

$client = new OpenRouterClient($apiKey);

$answer = $client->chat(
    [['role' => 'user', 'content' => 'What is the capital of France?']],
    'openai/gpt-4o-mini'
);

echo $answer; // plain string

The optional 4th parameter $options allows passing any API parameter without breaking existing calls:

$answer = $client->chat(
    [['role' => 'user', 'content' => 'Explain recursion briefly.']],
    'openai/gpt-4o',
    [],   // headers
    [
        'max_completion_tokens' => 256,
        'temperature'           => 0.5,
        'plugins'               => [['id' => 'web']],
    ]
);

4) Extended chat – chatEx()

Returns a full result array with content, usage, finish_reason, tool_calls, reasoning, and id.

$result = $client->chatEx(
    [['role' => 'user', 'content' => 'Summarize todays AI news.']],
    'openai/gpt-4o',
    ['plugins' => [['id' => 'web']]]
);

echo $result['content'];
echo $result['usage']['total_tokens'];

5) Tool calling

$result = $client->chatEx(
    [['role' => 'user', 'content' => "What's the weather in Berlin?"]],
    'openai/gpt-4o',
    [
        'tools' => [[
            'type'     => 'function',
            'function' => [
                'name'        => 'get_weather',
                'description' => 'Returns the current weather for a city',
                'parameters'  => [
                    'type'       => 'object',
                    'properties' => ['city' => ['type' => 'string']],
                    'required'   => ['city'],
                ],
            ],
        ]],
        'tool_choice' => 'auto',
    ]
);

if (!empty($result['tool_calls'])) {
    $fn   = $result['tool_calls'][0]['function']['name'];
    $args = json_decode($result['tool_calls'][0]['function']['arguments'], true);
}

6) Structured outputs

$data = $client->structuredChat(
    [['role' => 'user', 'content' => 'List 3 cities in Germany']],
    [
        'name'   => 'cities',
        'schema' => [
            'type'       => 'object',
            'properties' => [
                'cities' => ['type' => 'array', 'items' => ['type' => 'string']],
            ],
            'required' => ['cities'],
        ],
    ]
);

print_r($data['cities']); // ['Berlin', 'Hamburg', 'Munich']

7) Streaming

$client->stream(
    [['role' => 'user', 'content' => 'Write a short poem about the ocean.']],
    function(string $delta, bool $done): void {
        if ($done) { echo PHP_EOL; return; }
        echo $delta;
        flush();
    }
);

8) Reasoning models (Claude 3.7+, o1/o3)

$result = $client->chatEx(
    [['role' => 'user', 'content' => 'Solve this step by step: 3x + 7 = 22']],
    'anthropic/claude-3-7-sonnet',
    ['reasoning' => ['effort' => 'high']]
);

echo $result['reasoning']; // internal thinking text
echo $result['content'];   // final answer

9) Provider routing

$result = $client->chatEx(
    [['role' => 'user', 'content' => 'Hello']],
    'openai/gpt-4o',
    [
        'provider' => [
            'sort'            => 'price',
            'data_collection' => 'deny',
        ],
    ]
);

10) Generation stats (cost & token counts)

$result = $client->chatEx($messages, 'openai/gpt-4o');
$stats  = $client->getGenerationStats($result['id']);

echo 'Cost: $' . $stats['usage']['cost'];

Available options (chat() / chatEx())

Parameter	Type	Description
max_completion_tokens	int	Max tokens in the response
temperature	float	Creativity 0–2, default 1.0
top_p / top_k	float / int	Sampling parameters
frequency_penalty	float	Repetition penalty by frequency
presence_penalty	float	Repetition penalty by presence
stop	array	Stop sequences
seed	int	Reproducible output
tools	array	Tool / function definitions
tool_choice	string	auto | none | required
parallel_tool_calls	bool	Call multiple tools in parallel
response_format	array	json_object or json_schema
plugins	array	web, file-parser, response-healing, context-compression
provider	array	Routing: sort, allow, ignore, data_collection
reasoning	array	['effort' => 'low|medium|high']
modalities	array	text, audio, image
session_id	string	Session group for observability

Available models

• openai/gpt-4o-mini
• openai/gpt-4o
• anthropic/claude-3.5-sonnet
• anthropic/claude-3-7-sonnet
• google/gemini-2.5-flash
• mistralai/mistral-small
• x-ai/grok-3-mini
• … and 400+ more on OpenRouter