ineersa / symfony-ai-openai-chatgpt-platform

v0.1.0 2026-10-13 01:00 UTC

Requires

• php: >=8.5
• ext-openssl: *
• firebase/php-jwt: ^6.11 || ^7.0
• league/oauth2-client: ^2.8
• psr/log: ^3.0
• symfony/ai-open-responses-platform: ^0.14
• symfony/ai-platform: ^0.14
• symfony/cache: ^8.0
• symfony/clock: ^8.0
• symfony/console: ^8.1
• symfony/filesystem: ^8.0
• symfony/http-client: ^8.0
• symfony/lock: ^8.0
• symfony/process: ^8.0
• symfony/uid: ^8.0

Requires (Dev)

• friendsofphp/php-cs-fixer: ^3.94
• phpstan/phpstan: ^2.1
• phpstan/phpstan-phpunit: ^2.0
• phpunit/phpunit: ^13.2

Suggests

• ext-zstd: Lossless level-3 compression of Responses request bodies to mitigate gateway upload-size failures.

MIT

Published on Packagist.org on 2026-10-13 15:12 UTC

README

Use your ChatGPT subscription with Symfony AI through Sign in with ChatGPT. The package handles browser login, token refresh, and requests to the OpenAI Responses API using only HTTP and server-sent events (SSE).

Requires PHP 8.5+, Symfony AI 0.14, and a ChatGPT account that permits subscription sharing.

Warning

Install ext-zstd to enable request compression. Without it, uncompressed requests exceeding roughly 1 MiB can fail with HTTP 503 because of an observed OpenAI gateway upload-limit bug, not a documented universal API limit. Compression mitigates this bug but does not guarantee that the model's full context window will fit.

Install the package

composer require ineersa/symfony-ai-openai-chatgpt-platform

Log in and send a message

This basic chatgpt.php example provides a login command and streams a reply to "Hello". Add /var/chatgpt/ to .gitignore before logging in to keep credentials out of version control.

Warning

This package requires a non-empty prompt_cache_key, although the OpenAI API makes it optional. Reuse one key per conversation: the package sends it as prompt_cache_key, session-id, and x-client-request-id to aid routing and cache reuse for identical prompt prefixes, without guaranteeing cache hits.

<?php

require __DIR__.'/vendor/autoload.php';

use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Auth\AuthCommand;
use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Auth\AuthFileStore;
use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Auth\IdTokenVerifier;
use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Auth\OAuthConfig;
use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Auth\OAuthService;
use Symfony\AI\Platform\Bridge\OpenAIChatGPT\Factory;
use Symfony\AI\Platform\Bridge\OpenResponses\ModelCatalog;
use Symfony\AI\Platform\Bridge\OpenResponses\ResponsesModel;
use Symfony\AI\Platform\Capability;
use Symfony\AI\Platform\Message\Message;
use Symfony\AI\Platform\Message\MessageBag;
use Symfony\Component\Console\Application;
use Symfony\Component\Filesystem\Filesystem;
use Symfony\Component\HttpClient\HttpClient;
use Symfony\Component\Lock\LockFactory;
use Symfony\Component\Lock\Store\FlockStore;

// Keep credentials on disk and coordinate token refresh across local processes.
$lockDir = __DIR__.'/var/chatgpt/locks';
(new Filesystem())->mkdir($lockDir, 0700);
$http = HttpClient::create();
$store = new AuthFileStore(
	__DIR__.'/var/chatgpt/auth.json',
	new LockFactory(new FlockStore($lockDir)),
);
// Use your application's name for OAuth registration and the Originator header.
$config = new OAuthConfig('My CLI');
$oauth = new OAuthService($store, $http, new IdTokenVerifier($http), $config);

// Handle login, refresh, and disconnect through Symfony Console.
if ('ask' !== ($argv[1] ?? null)) {
	$console = new Application();
	$console->addCommand(new AuthCommand($oauth, $config));
	exit($console->run());
}

$modelName = $argv[2] ?? throw new InvalidArgumentException('Usage: php chatgpt.php ask MODEL');
$provider = Factory::createProvider($oauth, $http, new ModelCatalog());
// Each ask starts a new conversation. Persist this key with history to continue it.
$conversationKey = bin2hex(random_bytes(16));
// Invoke an account-supported model; OAuthService refreshes expired credentials.
$result = $provider->invoke(
	new ResponsesModel($modelName, [Capability::INPUT_MESSAGES, Capability::OUTPUT_TEXT]),
	new MessageBag(Message::ofUser('Hello')),
	['prompt_cache_key' => $conversationKey],
);
// Print text as SSE events arrive, rather than waiting for the complete response.
foreach ($result->asTextStream() as $text) {
	echo $text;
}
echo PHP_EOL;

Log in, then stream a reply. Replace YOUR_MODEL with a model available to your account:

php chatgpt.php auth:chatgpt login
php chatgpt.php ask YOUR_MODEL

If you signed in without enabling plan usage, run php chatgpt.php auth:chatgpt login --consent. The provider refreshes expired credentials before sending requests.

Documentation

Guide	Use it to
Authentication	Register the command, use manual login, refresh, and disconnect
Usage	Configure models, send conversation history, and cancel streams
History and reasoning	Preserve message phases, encrypted reasoning, and effort changes
API reference	Look up options, request limits, and error types
OAuth lifecycle	Understand verification, storage locks, and recovery after refresh

Develop

Run composer install, then castor test, castor lint, castor phpstan, castor cs-check, and castor composer-validate. Tests use mocked HTTP and synthetic credentials, not a real ChatGPT account.

MIT license.