A production-ready Laravel package to integrate with the Google Gemini API. Supports text, image, video, audio, long-context, structured output, function-calling and understanding capabilities.

#Laravel Gemini

A production-ready Laravel package to integrate with the Google Gemini API. Supports text, image, video, audio, long-context, structured output, function-calling and understanding capabilities.

#Features

• 🤖 Text generation with context and history
• 🖼️ Image generation and understanding
• 🎥 Video generation and analysis
• 🔊 Audio synthesis and transcription
• 📄 Document processing and understanding
• 🔍 Embeddings generation
• 📊 File management capabilities
• ⚡ Real-time streaming responses
• 🛡️ Configurable safety settings

#Installation

composer require hosseinhezami/laravel-gemini

Publish the configuration file:

php artisan vendor:publish --tag=gemini-config

Add your Gemini API key to your .env file:

GEMINI_API_KEY=your_gemini_api_key_here

#Configuration (detailed)

Configuration lives in config/gemini.php. Below are the most important keys and recommended defaults:

#Providers / models / methods

The providers array lets you map capability types to models and HTTP methods the provider uses:

Speech config (providers.gemini.default_speech_config) example:

'default_speech_config' => [
    'voiceName' => 'Kore',
    // 'speakerVoices' => [
    //     ['speaker' => 'Joe', 'voiceName' => 'Kore'],
    //     ['speaker' => 'Jane', 'voiceName' => 'Puck'],
    // ],
],

#Builder APIs — full method reference

This package exposes a set of builder-style facades: Gemini::text(), Gemini::image(), Gemini::video(), Gemini::audio(), Gemini::embeddings() and Gemini::files().

Below is a concise reference of commonly available chainable methods and what they do. Method availability depends on the builder.

#Common response helpers (Response object)

When you call ->generate() (or a polling save on long-running jobs) you typically get a response object with these helpers:

• content() — main textual output (string).
• model() — model name used.
• usage() — usage / billing info returned by the provider.
• requestId() — provider request id.
• save($path) — convenience method to download and persist a result to disk (media).

#Gemini::

use HosseinHezami\LaravelGemini\Facades\Gemini;

#TextBuilder (Gemini::text())

Use for: chat-like generation, long-context text, structured output, and multimodal understanding (text responses after uploading files).

Common methods:

Notes on history structure
History entries follow a role + parts format:

[
    ['role' => 'user', 'parts' => [['text' => 'User message']]],
    ['role' => 'model', 'parts' => [['text' => 'Assistant reply']]]
]

Text

$response = Gemini::text()
    ->model('gemini-2.5-flash')
    ->system('You are a helpful assistant.')
    ->prompt('Write a conversation between human and Ai')
    ->history([
        ['role' => 'user', 'parts' => [['text' => 'Hello AI']]],
        ['role' => 'model', 'parts' => [['text' => 'Hello human!']]]
    ])
    ->temperature( 0.7)
    ->maxTokens(1024)
    ->generate();
 
echo $response->content();

Streaming Responses

return response()->stream(function () use ($request) {
    Gemini::text()
        ->model('gemini-2.5-flash')
        ->prompt('Tell a long story about artificial intelligence.')
        ->stream(function ($chunk) {
            $text = $chunk['text'] ?? '';
            if (!empty(trim($text))) {
                echo "data: " . json_encode(['text' => $text]) . "\n\n";
                ob_flush();
                flush();
            }
        });
}, 200, [
    'Content-Type' => 'text/event-stream',
    'Cache-Control' => 'no-cache',
    'Connection' => 'keep-alive',
    'X-Accel-Buffering' => 'no',
]);

Document Understanding

$response = Gemini::text()
    ->upload('document', $filePath) // image, video, audio, document
    ->prompt('Extract the key points from this document.')
    ->generate();
 
echo $response->content();

Structured output

$response = Gemini::text()
    ->model('gemini-2.5-flash')
    ->structuredSchema([
    'type' => 'object',
    'properties' => [
        'name' => ['type' => 'string'],
        'age'  => ['type' => 'integer']
    ],
    'required' => ['name']
    ])
    ->prompt('Return a JSON object with name and age.')
    ->generate();
 
$json = $response->content(); // Parsable JSON matching the schema

#ImageBuilder (Gemini::image())

Use for image generation.

Image

$response = Gemini::image()
    ->model('gemini-2.5-flash-image-preview')
    ->method('generateContent')
    ->prompt('A futuristic city skyline at sunset.')
    ->generate();
 
$response->save('image.png');

#VideoBuilder (Gemini::video())

Use for short or long-running video generation.

Note: long-running video generation typically uses predictLongRunning or similar. The package abstracts polling & saving.

#AudioBuilder (Gemini::audio())

Use for TTS generation.

#Embeddings (Gemini::embeddings())

Accepts a payload array. Typical shape:

$embeddings = Gemini::embeddings([
    'model' => 'gemini-embedding-001',
    'content' => ['parts' => [['text' => 'Text to embed']]],
]);
 
/* embedding_config */
// https://ai.google.dev/gemini-api/docs/embeddings
// 'embedding_config': {
//     'task_type': 'SEMANTIC_SIMILARITY', // SEMANTIC_SIMILARITY, CLASSIFICATION, CLUSTERING, RETRIEVAL_DOCUMENT, RETRIEVAL_QUERY, CODE_RETRIEVAL_QUERY, QUESTION_ANSWERING, FACT_VERIFICATION
//     'output_dimensionality': 768 // 128, 256, 512, 768, 1536, 2048
// }

Return value is the raw embeddings structure (provider-specific). Use these vectors for semantic search, similarity, clustering, etc.

#Files API (Gemini::files())

High level file manager for uploads used by the "understanding" endpoints.

Files

// Upload a file
$uri = Gemini::files()->upload('document', $pathToFile);
 
// List all files
$files = Gemini::files()->list();
 
// Get file details
$fileInfo = Gemini::files()->get($file_id);
 
// Delete a file
$success = Gemini::files()->delete($file_id);

Supported file types & MIME

#Streaming (Server-Sent Events)

The stream route uses Content-Type: text/event-stream. Connect from a browser or SSE client and consume data: <json> messages per chunk.

#Streaming behaviour

• Implemented using SSE (Server-Sent Events). The stream yields chunks where each chunk is typically ['text' => '...'].
• Client should reconnect behaviorally for resilience and handle partial chunks.
• Use response headers:
  • Content-Type: text/event-stream
  • Cache-Control: no-cache
  • Connection: keep-alive
  • X-Accel-Buffering: no

#Tips, error handling & best practices

• Respect provider limits — pick appropriate maxTokens and temperature.
• For large media (video) prefer long-running predictLongRunning models and rely on save() to poll and download final asset.
• Use safetySettings from config for content filtering. You can override per-request.
• When uploading user-supplied files, validate MIME type and size before calling Gemini::files()->upload.

#Artisan Commands

The package includes helpful Artisan commands:

#License

This package is open-source software licensed under the MIT license.

#Contributing

Contributions, bug reports and pull requests are welcome.