Integrating OpenAI GPT-4 into Laravel: Beyond the Hello World

Everyone's tutorial shows you how to send a prompt and get a response. Here's how to do it properly in production.

Setup

composer require openai-php/client\n# .env\nOPENAI_API_KEY=sk-...\nOPENAI_ORGANIZATION=org-...

The Service Class Pattern

Never call the OpenAI client directly from controllers. Wrap it in a service:

class AiService {\n    private Client $client;\n\n    public function __construct() {\n        $this->client = OpenAI::client(config('services.openai.key'));\n    }\n\n    public function complete(string $prompt, array $options = []): string {\n        $response = $this->client->chat()->create(array_merge([\n            'model' => 'gpt-4o',\n            'messages' => [['role' => 'user', 'content' => $prompt]],\n            'max_tokens' => 1000,\n        ], $options));\n\n        return $response->choices[0]->message->content;\n    }\n}

Streaming Responses

For long responses, streaming dramatically improves UX. Users see output immediately instead of waiting 10+ seconds:

public function stream(string $prompt): \Generator {\n    $stream = $this->client->chat()->createStreamed([\n        'model' => 'gpt-4o',\n        'messages' => [['role' => 'user', 'content' => $prompt]],\n    ]);\n\n    foreach ($stream as $chunk) {\n        $content = $chunk->choices[0]->delta->content;\n        if ($content) yield $content;\n    }\n}

Cost Control (Critical in Production)

OpenAI costs can spiral fast. Here's how I keep them in check:

1. Cache responses — If users ask similar questions, cache the AI response in Redis for 24h
2. Rate limit per user — Use Laravel's built-in rate limiting on your AI endpoints
3. Log all API calls — Store prompt tokens, completion tokens, and cost per call in the database
4. Set model-appropriate defaults — Use gpt-4o-mini for simple tasks, gpt-4o only when quality matters

Prompt Engineering Tips

Your system prompt is your most powerful lever. Be explicit about output format, length, and tone. Use JSON mode for structured outputs.