Vertex AI

AnthropicVertexClient runs the shared Anthropic completion adapter through Anthropic’s official Vertex SDK.

import { AnthropicVertexClient } from '@anvia/anthropic'

const vertex = new AnthropicVertexClient({
  projectId: 'my-gcp-project',
  region: 'global',
})

const model = vertex.completionModel('claude-sonnet-5')

Authentication

The official SDK follows Google authentication conventions. Environment-based setup can use:

export ANTHROPIC_VERTEX_PROJECT_ID='my-gcp-project'
export CLOUD_ML_REGION='global'
export GOOGLE_APPLICATION_CREDENTIALS='/absolute/path/service-account.json'

Application Default Credentials are preferable to embedding service-account JSON in application configuration. Never commit credential files.

Custom authentication options from the official SDK are also accepted:

const vertex = new AnthropicVertexClient({
  projectId: 'my-gcp-project',
  region: 'global',
  googleAuth,
})

The option type also permits an already configured AnthropicVertex client.

Behavioral differences

• completionModel() defaults to claude-sonnet-5, not the standard client’s older default.
• The returned completion model has the same Anvia request and stream contracts.
• AnthropicVertexClient does not implement listModels() because Vertex does not expose Anthropic’s Models API through this client.
• Available IDs, regions, quotas, and authentication failures are governed by the Vertex deployment.

Production considerations

• Validate the project and region at startup.
• Give the workload identity only required Vertex permissions.
• Avoid passing externally supplied credential objects without validation.
• Record project/region as safe deployment metadata, not secrets.
• Test streaming and tool use in the target region.
• Do not assume an Anthropic API model ID is available through Vertex under the same name.