Google AI

The google-ai provider supports Google's Gemini models via the Generative Language API.

tip

BAML automatically selects streamGenerateContent when using the streaming interface.

Basic Example

client<llm> MyClient {
    provider google-ai
    options {
        model "gemini-2.5-flash"
    }
}

BAML-Specific Options

Option	Type	Default	Description
api_key	string	env.GOOGLE_API_KEY	API key
base_url	string	https://generativelanguage.googleapis.com/v1beta	Base URL
model	string	gemini-2.5-flash	Model identifier
headers	object	-	Additional headers

Available Models

Model	Use Case	Context	Features
gemini-2.5-pro	Complex tasks, coding, STEM	1M	Adaptive thinking, multimodal
gemini-2.5-flash	Production apps, balanced	1M	Best price/performance
gemini-2.5-flash-lite	High-volume, cost-sensitive	1M	Lowest cost, fastest

See the Google Model Docs for the latest models.

Configuration Example

tip

Parameters like temperature for Gemini models are specified in the generationConfig object.

client<llm> MyClient {
    provider google-ai
    options {
        model "gemini-2.5-flash"
        headers {
            "X-My-Header" "my-value"
        }
        generationConfig {
            temperature 0.5
        }
    }
}

Provider Parameters

Option	Type	Description
generationConfig	object	Generation settings (temperature, maxOutputTokens, etc.)
safetySettings	array	Safety filter configuration

note

Google AI uses send_base64_unless_google_url by default for images, preserving Google Cloud Storage URLs (gs://) while converting other URLs to base64.

See the Google Gemini API documentation for all options.

See Also

• Vertex AI (for enterprise Google Cloud)
• Client Overview