Use the OpenAI API in Laravel

1. Install the client
2. Store your API key
3. Make your first call
4. Stream the response
5. Handle errors and limits
6. Or run it locally

Wiring an LLM into a Laravel app is less work than most people expect. The openai-php/laravel package gives you a typed client and a facade, so calling the chat endpoint is a single method call. Here's a clean setup, from installing the client to streaming responses and keeping your key safe.

#Install the client

Pull in the package with Composer:

composer require openai-php/laravel

Then publish its config file:

php artisan vendor:publish --provider="OpenAI\Laravel\ServiceProvider"

That creates config/openai.php, which reads your credentials from the environment.

#Store your API key

Put your key in .env — never in code or version control:

OPENAI_API_KEY=sk-your-key-here

The published config already references it, so there's nothing else to wire up:

// config/openai.php
'api_key' => env('OPENAI_API_KEY'),

On a production server, set OPENAI_API_KEY as a real environment variable or a secret rather than shipping the .env. If the key ever leaks, rotate it in your OpenAI dashboard immediately.

#Make your first call

With the facade, a chat completion is one call:

use OpenAI\Laravel\Facades\OpenAI;

$response = OpenAI::chat()->create([
    'model' => 'gpt-4o-mini',
    'messages' => [
        ['role' => 'system', 'content' => 'You are a concise assistant.'],
        ['role' => 'user', 'content' => 'Summarise what a reverse proxy does.'],
    ],
    'max_tokens' => 200,
]);

return $response->choices[0]->message->content;

Picking a small model like gpt-4o-mini and capping max_tokens are the two cheapest habits for keeping your bill predictable.

#Stream the response

For a chat UI you want text to appear as it's generated rather than after a long pause. Use the streaming variant and emit each chunk:

use OpenAI\Laravel\Facades\OpenAI;

return response()->stream(function () {
    $stream = OpenAI::chat()->createStreamed([
        'model' => 'gpt-4o-mini',
        'messages' => [['role' => 'user', 'content' => 'Write a haiku about servers.']],
    ]);

    foreach ($stream as $chunk) {
        echo $chunk->choices[0]->delta->content ?? '';
        ob_flush();
        flush();
    }
}, 200, [
    'Content-Type' => 'text/event-stream',
    'X-Accel-Buffering' => 'no',
]);

The X-Accel-Buffering: no header matters behind Nginx, which otherwise buffers the whole response and defeats the point of streaming.

#Handle errors and limits

Two failures are worth handling explicitly. The first is the rate limit — under load OpenAI returns HTTP 429, and you should back off and retry rather than hammer the endpoint; the dedicated guide is fix the OpenAI API rate limit (429) error. The second is cost: set a hard spending limit in your account, cache repeated prompts, and reach for the smallest model that does the job.

#Or run it locally

If you'd rather not depend on a paid API at all, the same code can point at a local model. Ollama exposes an OpenAI-compatible endpoint, so changing the base URL to http://localhost:11434/v1 lets this exact Laravel setup talk to a model running on your own server.