README

A Laravel AI provider that proxies Gemini CLI and Antigravity endpoints to access Google's CloudCode-PA v1internal API. Use Gemini, Claude, and GPT models through a single unified interface with automatic rate-limit cascading, multi-endpoint fallback, and SSE streaming.

Warning

Use at Your Own Risk

This package accesses Google's internal CloudCode-PA API through undocumented endpoints using OAuth credentials intended for Gemini CLI. This is not an officially supported method of accessing these models. Using this package may violate Google's Terms of Service and could result in your Google account being suspended or banned. The authors accept no responsibility for any consequences arising from its use.

Requirements

PHP 8.3+
Laravel 12 or 13
Laravel AI ^0.1 | ^0.2 (pre-release — expect breaking changes)
OAuth credentials JSON produced by Gemini CLI

Installation

composer require ursamajeur/laravel-ai-gemini-driver

The service provider is auto-discovered — no manual registration needed.

Publish the configuration file:

php artisan vendor:publish --tag=cloudcode-pa-config

Configuration

Authentication

This package requires an OAuth credentials JSON file produced by Gemini CLI. Run Gemini CLI at least once and authenticate — this generates the OAuth JSON that this package uses. The package handles token refresh automatically after that.

Set the path to your OAuth credentials in .env:

CLOUDCODE_PA_CREDENTIALS_PATH=storage/cloudcode-pa/oauth_creds.json

If not set, it defaults to storage/cloudcode-pa/oauth_creds.json. The package handles token refresh automatically.

Project ID

The CloudCode-PA project ID is obtained during Gemini CLI onboarding. If not configured, the package will auto-discover it via the loadCodeAssist RPC.

CLOUDCODE_PA_PROJECT=your-project-id

Models

Configure default models via environment variables:

CLOUDCODE_PA_DEFAULT_MODEL=claude-opus-4
CLOUDCODE_PA_CHEAPEST_MODEL=gemini-2.5-flash
CLOUDCODE_PA_SMARTEST_MODEL=claude-opus-4
CLOUDCODE_PA_RERANKING_MODEL=gemini-2.5-flash

Supported Models

Partner Models

Model	Alias
Claude Opus 4	`claude-opus-4`
Claude Sonnet 4.5	`claude-sonnet-4-5`
Claude Sonnet 4	`claude-sonnet-4`
Claude Haiku 4.5	`claude-haiku-4-5`
GPT-4.1	`gpt-4-1`
GPT OSS 120B	`gpt-oss-120b-medium`

Gemini Models

Model	Alias
Gemini 3 Pro (preview)	`gemini-3-pro-preview`
Gemini 3 Flash (preview)	`gemini-3-flash-preview`
Gemini 2.5 Pro	`gemini-2.5-pro`
Gemini 2.5 Flash	`gemini-2.5-flash`
Gemini 2.5 Flash Lite	`gemini-2.5-flash-lite`
Gemini 2.0 Flash	`gemini-2.0-flash`

You can add custom model aliases in config/cloudcode-pa.php under the models key.

Usage

The provider name is cloudcode-pa. Use it with Laravel AI's standard API:

use Laravel\Ai\Ai;

// Text generation
$provider = Ai::textProvider('cloudcode-pa');

// Reranking
$provider = Ai::rerankingProvider('cloudcode-pa');

For the full API (text generation, streaming, multi-turn conversations, reranking), refer to the Laravel AI documentation.

Key Features

Model Cascade (Rate-Limit Fallback)

When a model returns 429 (Rate Limited), the gateway automatically cascades through a configurable fallback chain. Each step tries all configured endpoints before advancing to the next model.

Default cascade: claude-opus-4 → gemini-3-pro-preview → gemini-3-flash-preview → gemini-2.5-pro → gemini-2.5-flash

Cascade only activates when using the default model — explicit model requests use endpoint fallback only.

CLOUDCODE_PA_CASCADE_ENABLED=true

Customize cascade steps in config/cloudcode-pa.php:

'cascade' => [
    'enabled' => true,
    'steps' => [
        'claude-opus-4',
        'gemini-2.5-pro',
        'gemini-2.5-flash',
    ],
],

Multi-Endpoint Fallback

Multiple endpoints with independent quota pools are configured by default. On endpoint failure or rate limiting, the next endpoint is tried automatically. Endpoints are customizable in config/cloudcode-pa.php.

Automatic Model Routing

Partner models (Claude, GPT) and Gemini models use different API protocols. The package detects the model type and routes automatically — no configuration needed.

Timeouts

CLOUDCODE_PA_TIMEOUT=30            # Non-streaming requests (seconds)
CLOUDCODE_PA_STREAM_TIMEOUT=120    # SSE streaming requests (seconds)
CLOUDCODE_PA_CONNECT_TIMEOUT=10    # Connection timeout (seconds)

Debug Logging

Enable request/response logging (credentials are always redacted):

CLOUDCODE_PA_DEBUG=true

Testing

composer test

License

MIT

ursamajeur / laravel-ai-gemini-driver

Maintainers

Package info

Statistics

Security