Skip to main content
By default, all Vertex AI providers share the same configuration from the vertex block in your config/prism.php. However, you can override configuration for specific providers by adding per-provider config blocks.

​
How provider overrides work

When you use a specific provider constant (like Vertex::Anthropic or Vertex::Mistral), Prism checks for a provider-specific config block that matches the provider key. If found, it merges those settings with the base vertex config, with the provider-specific settings taking precedence.

​
Provider keys

Each provider constant has a corresponding config key:
ConstantConfig Key
Vertex::Geminivertex-gemini
Vertex::Anthropicvertex-anthropic
Vertex::Mistralvertex-mistral
Vertex::Metavertex-meta
Vertex::DeepSeekvertex-deepseek
Vertex::AI21vertex-ai21
Vertex::Kimivertex-kimi
Vertex::MiniMaxvertex-minimax
Vertex::OpenAIvertex-openai
Vertex::Qwenvertex-qwen
Vertex::ZAIvertex-zai

​
Common override scenarios

​
Different regions for different providers

You might want to use different Vertex AI regions for different providers (for example, to take advantage of regional model availability or lower latency): 
// config/prism.php
'providers' => [
    'vertex' => [
        'project_id'  => env('VERTEX_PROJECT_ID'),
        'location'    => env('VERTEX_LOCATION', 'us-central1'),
        'credentials' => env('VERTEX_CREDENTIALS'),
    ],

    // Use Europe region for Anthropic models
    'vertex-anthropic' => [
        'location' => 'europe-west1',
    ],

    // Use Asia region for Gemini models
    'vertex-gemini' => [
        'location' => 'asia-northeast1',
    ],
],
Now when you use Vertex::Anthropic, it will use europe-west1, and Vertex::Gemini will use asia-northeast1. All other providers will use the default us-central1.

​
Separate credentials per provider

You can use different service accounts or API keys for different providers: 
// config/prism.php
'providers' => [
    'vertex' => [
        'project_id'  => env('VERTEX_PROJECT_ID'),
        'location'    => 'us-central1',
        'credentials' => env('VERTEX_CREDENTIALS'),
    ],

    // Use different service account for Anthropic
    'vertex-anthropic' => [
        'credentials' => env('VERTEX_ANTHROPIC_CREDENTIALS'),
    ],

    // Use API key for Gemini (Express mode)
    'vertex-gemini' => [
        'api_key' => env('VERTEX_GEMINI_API_KEY'),
        'project_id' => null,
        'location' => null,
    ],
],
When you set project_id and location to null for a specific provider, that provider will use Express mode even if the base vertex config has Standard mode settings.

​
Different projects per provider

You can route different providers to different Google Cloud projects: 
// config/prism.php
'providers' => [
    'vertex' => [
        'project_id'  => env('VERTEX_PROJECT_ID'),
        'location'    => 'us-central1',
        'credentials' => env('VERTEX_CREDENTIALS'),
    ],

    // Use different project for Mistral models
    'vertex-mistral' => [
        'project_id' => env('VERTEX_MISTRAL_PROJECT_ID'),
        'credentials' => env('VERTEX_MISTRAL_CREDENTIALS'),
    ],
],

​
Override precedence

The merge logic works as follows:
  1. Start with the base vertex config
  2. Merge in the provider-specific config (e.g., vertex-anthropic)
  3. Provider-specific values override base values
Example: 
'providers' => [
    'vertex' => [
        'project_id'  => 'my-project',
        'location'    => 'us-central1',
        'credentials' => '/path/to/default.json',
        'api_key'     => null,
    ],

    'vertex-anthropic' => [
        'location'    => 'europe-west1',  // Overrides base location
        'credentials' => '/path/to/anthropic.json',  // Overrides base credentials
        // project_id is inherited from base config
    ],
],
When using Vertex::Anthropic, the final config will be: 
[
    'project_id'  => 'my-project',  // from base
    'location'    => 'europe-west1',  // overridden
    'credentials' => '/path/to/anthropic.json',  // overridden
    'api_key'     => null,  // from base
]

​
Mixing Standard and Express modes

You can use Standard mode for some providers and Express mode for others: 
// config/prism.php
'providers' => [
    // Base config uses Standard mode
    'vertex' => [
        'project_id'  => env('VERTEX_PROJECT_ID'),
        'location'    => env('VERTEX_LOCATION', 'us-central1'),
        'credentials' => env('VERTEX_CREDENTIALS'),
    ],

    // Override Gemini to use Express mode with API key
    'vertex-gemini' => [
        'api_key' => env('VERTEX_GEMINI_API_KEY'),
        'project_id' => null,
        'location' => null,
        'credentials' => null,
    ],
],
Remember that Express mode only supports Gemini models. You cannot use Express mode for Anthropic, Mistral, Meta, or other partner providers.

​
Runtime provider options

In addition to config-based overrides, you can pass provider-specific options at runtime using withProviderOptions(): 
use Prism\Prism\Prism;
use Prism\Vertex\Enums\Vertex;
use Prism\Vertex\Enums\VertexSchema;

$response = Prism::text()
    ->using(Vertex::Gemini, 'gemini-2.5-flash')
    ->withProviderOptions([
        'apiSchema' => VertexSchema::Gemini,
        'safetySettings' => [
            ['category' => 'HARM_CATEGORY_HATE_SPEECH', 'threshold' => 'BLOCK_NONE'],
        ],
        'thinkingBudget' => 8192,
        'includeThoughts' => true,
    ])
    ->withPrompt('Explain quantum computing')
    ->asText();
These options are request-specific and do not affect the provider configuration.

​
Available provider options

  • apiSchema - Override the API schema (VertexSchema enum)
  • safetySettings - Array of safety threshold configs
  • thinkingBudget - Max thinking tokens for reasoning models
  • thinkingLevel - Thinking intensity level
  • includeThoughts - Include thinking process in response
  • thinkingConfig - Full thinking configuration object

​
Validation

Prism Vertex validates your configuration when the provider is instantiated. You’ll receive a PrismException if:
  • You use Standard mode without project_id and location
  • You use Express mode without api_key
  • You use Express mode with a non-Gemini schema
  • You specify credentials but the file doesn’t exist
// This will throw an exception:
Prism::text()
    ->using(Vertex::Anthropic, 'claude-3-5-sonnet@20241022')
    ->withPrompt('Hello')
    ->asText();

// If vertex-anthropic config has:
'vertex-anthropic' => [
    'api_key' => env('VERTEX_API_KEY'),
    'project_id' => null,
    'location' => null,
]

// Error: "Vertex AI Express mode only supports Google models.
// The anthropic apiSchema requires Standard mode with project_id and location."

​
Environment-specific overrides

You can use Laravel’s environment-specific config files to define different overrides per environment: 
// config/prism.php (production)
'providers' => [
    'vertex' => [
        'project_id'  => env('VERTEX_PROJECT_ID'),
        'location'    => 'us-central1',
        'credentials' => env('VERTEX_CREDENTIALS'),
    ],
],

// config/testing/prism.php (testing)
'providers' => [
    'vertex' => [
        'project_id'  => env('VERTEX_TEST_PROJECT_ID'),
        'location'    => 'us-central1',
        'credentials' => env('VERTEX_TEST_CREDENTIALS'),
    ],

    'vertex-gemini' => [
        'api_key' => env('VERTEX_TEST_API_KEY'),
        'project_id' => null,
        'location' => null,
    ],
],

Build docs developers (and LLMs) love

Get started for freeTalk to us