orbiter.models.vertex

Google Vertex AI LLM provider implementation. Wraps the google-genai SDK with Vertex AI (GCP Application Default Credentials) authentication to implement ModelProvider.complete() and ModelProvider.stream() with normalized response types.

Module Path

from orbiter.models.vertex import VertexProvider

Auto-Registration

On import, VertexProvider is registered in model_registry under the name "vertex":

model_registry.register("vertex", VertexProvider)

VertexProvider

Wraps the google.genai.Client with Vertex AI authentication for GCP-hosted model access. Supports Gemini models and Model Garden models available through Vertex AI.

Inherits: ModelProvider

Constructor

VertexProvider(config: ModelConfig)

The constructor creates a genai.Client with vertexai=True using:

• project from the GOOGLE_CLOUD_PROJECT environment variable (defaults to "")
• location from the GOOGLE_CLOUD_LOCATION environment variable (defaults to "us-central1")

Authentication is handled via Application Default Credentials (ADC) — no API key is needed.

Environment Variables

Methods

complete()

async def complete(
    self,
    messages: list[Message],
    *,
    tools: list[dict[str, Any]] | None = None,
    temperature: float | None = None,
    max_tokens: int | None = None,
) -> ModelResponse

Send a completion request to Vertex AI.

Returns: ModelResponse

Raises: ModelError if the API call fails.

stream()

async def stream(
    self,
    messages: list[Message],
    *,
    tools: list[dict[str, Any]] | None = None,
    temperature: float | None = None,
    max_tokens: int | None = None,
) -> AsyncIterator[StreamChunk]

Stream a completion from Vertex AI. Uses generate_content_stream() internally.

Yields: StreamChunk

Raises: ModelError if the API call fails.

Message Conversion

Identical to the Gemini provider — messages are converted to Google API format:

Tool Schema Conversion

Same as the Gemini provider — tools are converted from OpenAI format to Google’s function_declarations format.

Finish Reason Mapping

Same as the Gemini provider.

Error Prefix

Vertex AI errors use the "vertex:" prefix in ModelError messages (e.g., "vertex:gemini-2.0-flash"), distinguishing them from Gemini API errors which use the "gemini:" prefix.

Example

import asyncio
from orbiter.models import get_provider
from orbiter.types import SystemMessage, UserMessage

async def main():
    # No API key needed -- uses Application Default Credentials
    provider = get_provider("vertex:gemini-2.0-flash")

    response = await provider.complete(
        [
            SystemMessage(content="You are a helpful assistant."),
            UserMessage(content="What is the capital of France?"),
        ],
        temperature=0.0,
        max_tokens=100,
    )

    print(response.content)  # "The capital of France is Paris."
    print(response.usage)    # Usage(input_tokens=25, output_tokens=8, total_tokens=33)

asyncio.run(main())

Authentication Setup

Vertex AI uses Application Default Credentials (ADC). Set up before using the provider:

# Option 1: gcloud CLI (development)
gcloud auth application-default login
export GOOGLE_CLOUD_PROJECT="my-project-id"
export GOOGLE_CLOUD_LOCATION="us-central1"

# Option 2: Service account (production)
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/service-account.json"
export GOOGLE_CLOUD_PROJECT="my-project-id"

Supported Models

Any model available through Vertex AI, including:

• Gemini models: gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash
• Model Garden: Third-party and open models hosted on Vertex AI

Gemini API vs Vertex AI

For API-key-based access, use the Gemini provider instead.