README

Prism Vertex

A standalone Google Vertex AI provider for the Prism PHP framework. Access Google Gemini and partner model families through a single configuration and unified interface.

Installation

composer require rmh/vertex

Configuration

Add the following to your Prism configuration (config/prism.php):

Standard mode (project + location required)

'vertex' => [
    'project_id'  => env('VERTEX_PROJECT_ID'),
    'location'    => env('VERTEX_LOCATION', 'us-central1'),
    'credentials' => env('VERTEX_CREDENTIALS'), // path to service-account.json
    // OR use api_key instead of credentials:
    // 'api_key' => env('VERTEX_API_KEY'),
],

Express mode (API key only, Google models only)

When project_id and location are omitted, the package automatically uses Vertex AI Express Mode endpoints:

'vertex' => [
    'api_key' => env('VERTEX_API_KEY'),
    // project_id and location omitted — triggers Express mode
],

Important

Express mode only supports Google Gemini models. Using partner model providers in Express mode will throw an exception.

Usage

Use the provider constants from Prism\Vertex\Enums\Vertex with the Prism facade:

Text Generation

use Prism\Prism\Prism;
use Prism\Vertex\Enums\Vertex;

// Google Gemini
$response = Prism::text()
    ->using(Vertex::Gemini, 'gemini-2.5-flash')
    ->withPrompt('Explain quantum computing in simple terms')
    ->asText();

echo $response->text;

// Anthropic Claude on Vertex AI
$response = Prism::text()
    ->using(Vertex::Anthropic, 'claude-3-5-sonnet@20241022')
    ->withPrompt('Explain quantum computing in simple terms')
    ->asText();

// Meta Llama on Vertex AI
$response = Prism::text()
    ->using(Vertex::Meta, 'llama-4-scout-17b-16e-instruct-maas')
    ->withPrompt('Explain quantum computing in simple terms')
    ->asText();

Structured Output (JSON)

use Prism\Prism\Prism;
use Prism\Vertex\Enums\Vertex;
use Prism\Prism\Schema\ObjectSchema;
use Prism\Prism\Schema\StringSchema;
use Prism\Prism\Schema\ArraySchema;

$schema = new ObjectSchema(
    name: 'languages',
    description: 'Top programming languages',
    properties: [
        new ArraySchema(
            'languages',
            'List of programming languages',
            items: new ObjectSchema(
                name: 'language',
                description: 'Programming language details',
                properties: [
                    new StringSchema('name', 'The language name'),
                    new StringSchema('popularity', 'Popularity description'),
                ]
            )
        )
    ]
);

$response = Prism::structured()
    ->using(Vertex::Gemini, 'gemini-2.5-flash')
    ->withSchema($schema)
    ->withPrompt('List the top 3 programming languages')
    ->asStructured();

$data = $response->structured;

Embeddings

use Prism\Prism\Prism;
use Prism\Vertex\Enums\Vertex;

$response = Prism::embeddings()
    ->using(Vertex::Gemini, 'text-embedding-005')
    ->fromInput('The sky is blue')
    ->asEmbeddings();

$embeddings = $response->embeddings;

Streaming

Warning

Streaming is not yet supported. This feature is planned for a future release.

Supported Providers

All providers share one vertex config block. The provider constant determines which API schema and publisher endpoint is used:

API Schemas

Prism Vertex uses three API schemas to handle the different formats used by Vertex AI publishers:

• Gemini — Google's native generateContent / predict endpoints.
• Anthropic — Uses :rawPredict with the Anthropic Messages API format (anthropic_version: vertex-2023-10-16).
• OpenAI — Uses :rawPredict or :chatCompletions with OpenAI-compatible format for partner models (Mistral, Meta, DeepSeek, AI21, Kimi, MiniMax, OpenAI-OSS, Qwen, ZAI).

The provider constant automatically selects the correct schema. You can override it via withProviderOptions():

use Prism\Vertex\Enums\VertexSchema;

$response = Prism::text()
    ->using(Vertex::Gemini, 'some-model')
    ->withProviderOptions(['apiSchema' => VertexSchema::Anthropic])
    ->withPrompt('Hello')
    ->asText();

Structured Output Details

Each schema handles structured output differently:

• Gemini — Native structured output via response_mime_type: application/json and response_schema. The model is constrained to produce valid JSON matching your schema.
• OpenAI — Uses response_format: { type: "json_object" } combined with a schema instruction message. All MaaS partner models support this mode.
• Anthropic — No native JSON mode. A prompt is appended instructing the model to respond with JSON conforming to the provided schema.

You can override the schema instruction prompt for Anthropic and OpenAI schemas using withProviderOptions():

Prism::structured()
    ->using(Vertex::Anthropic, 'claude-3-5-sonnet@20241022')
    ->withSchema($schema)
    ->withProviderOptions([
        'jsonModeMessage' => 'My custom JSON instruction message',
    ])
    ->withPrompt('My prompt')
    ->asStructured();

Authentication

Service Account (recommended for production)

Set the credentials config option to the path of your service account JSON key file. The package uses google/auth to obtain a Bearer token automatically.

API Key

Set the api_key config option. The key is sent as a ?key= query parameter on every request. This works for both Standard and Express modes.

Per-Provider Config Overrides

All providers read from the shared vertex config. If you need different credentials for a specific provider, add a per-provider config block — it will override the shared config:

'vertex' => [
    'project_id'  => env('VERTEX_PROJECT_ID'),
    'location'    => env('VERTEX_LOCATION', 'us-central1'),
    'credentials' => env('VERTEX_CREDENTIALS'),
],

// Override for Anthropic only (e.g. different region)
'vertex-anthropic' => [
    'location' => 'europe-west1',
],

Requirements

• PHP 8.2+
• Laravel 11 or 12
• prism-php/prism >= 0.99.7

License

The MIT License (MIT). Please see License File for more information.