Langfuse OpenAI Middleware for PHP

Introduction

This library provides wraper functions to use Langfuse LLM monitoring for your application. It was build for Symfony but can be used in any PHP application.

Installation

Install the library and required dependencies via Composer:

composer require janzaba/langfuse

Configuration in Symfony

Step 1: Define Environment Variables

In your .env file, add your Langfuse PUBLIC_KEY and SECRET_KEY:

LANGFUSE_PUBLIC_KEY=your-public-key
LANGFUSE_SECRET_KEY=your-secret-key

Step 2: Register Services

In your config/services.yaml, add the following service definitions:

parameters:
    langfuse_config:
        public_key: '%env(LANGFUSE_PUBLIC_KEY)%'
        secret_key: '%env(LANGFUSE_SECRET_KEY)%'
        # Optional: langfuse_base_uri: 'https://custom.langfuse.endpoint/'

services:
        Langfuse\Config\Config:
        class: Langfuse\Config\Config
        arguments:
            - '%langfuse_config%'
        public: false

    Langfuse\Client\LangfuseClient:
        arguments:
            $config: '@Langfuse\Config\Config'

    Langfuse\LangfuseManager:
        arguments:
            $langfuseClient: '@Langfuse\Client\LangfuseClient'

Step 3: Use the OpenAI Client in Your Services

Now you can wrap your code with helper methods

Trace

$this->langfuseManager->withTrace(
    'Trace name',
    ['operation' => 'example operation name'],
    function () {
        // Your code here
    }
);

Generation

Inside a trace you can have LLM generation.

$answer = $this->langfuseManager->withGeneration(
    'prompt name', 
    'gpt-4o-mini', 
    $prompt, 
    function () use ($prompt) {
        return $this->openAIClient->chat()->create(
            [
                'model' => 'gpt-4o-mini',
                'messages' => $prompt,
            ]
        );
    }
);

Prompt Management

You can retrieve prompts from Langfuse to manage your prompts centrally:

// Get production prompt (default)
$prompt = $this->langfuseManager->getPrompt("Extract claims from text");

// Get by label
$prompt = $this->langfuseManager->getPrompt("Extract claims from text", "production");
$prompt = $this->langfuseManager->getPrompt("Extract claims from text", "latest");

// Get by version (not recommended as it requires code changes to deploy new prompt versions)
$prompt = $this->langfuseManager->getPrompt("Extract claims from text", null, 1);

// Use the prompt
$promptText = $prompt->getPrompt();
$promptVersion = $prompt->getVersion();

// Compile prompt with variables
$compiledPrompt = $prompt->compile([
    'topic' => 'climate change',
    'max_words' => 100
]);

// Integrate with generation tracking
$answer = $this->langfuseManager->withGeneration(
    'summarize-article', 
    'gpt-4o-mini', 
    ['role' => 'system', 'content' => $compiledPrompt], 
    function () use ($compiledPrompt) {
        return $this->openAIClient->chat()->create([
            'model' => 'gpt-4o-mini',
            'messages' => [
                ['role' => 'system', 'content' => $compiledPrompt]
            ],
        ]);
    }
);

Contributing

Contributions are welcome! Please submit a pull request or open an issue for any improvements or bugs.

License

This project is licensed under the MIT License.