Summary

Add support for Google Vertex AI as a model provider, using Application Default Credentials (ADC) for authentication.

Motivation

• Users with GCP accounts and free credits ($300 trial) would like to use Vertex AI
• Vertex AI offers features not available in Google AI Studio:
  • Grounding (search-augmented generation)
  • Higher rate limits
  • Context Caching for cost optimization
  • Enterprise SLA and support
• ADC is the standard GCP authentication method and works seamlessly with gcloud auth application-default login

Proposed Implementation

Add a new provider configuration option, for example:

{
  "models": {
    "providers": {
      "vertex-ai": {
        "api": "google-vertex",
        "project": "my-gcp-project",
        "location": "us-central1",
        "models": [
          {
            "id": "gemini-2.0-flash",
            "name": "Vertex Gemini 2.0 Flash"
          }
        ]
      }
    }
  }
}

Authentication should use ADC (via google-auth-library or similar), reading credentials from:

1. GOOGLE_APPLICATION_CREDENTIALS environment variable
2. Default credential chain (~/.config/gcloud/application_default_credentials.json)

Additional Context

• Current google-generative-ai API works great for AI Studio, but requires API keys
• Vertex AI uses a different endpoint format: https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/publishers/google/models/{model}
• The Gemini models available on Vertex AI are the same as AI Studio, just with different auth and endpoints

Thanks for considering!