Configuration

GeminiClientOptions prevents mixing Gemini API-key and Vertex settings.

Gemini Developer API

const gemini = new GeminiClient({
  apiKey: process.env.GEMINI_API_KEY,
})

apiKey is required unless client is supplied. Do not set project or location in this mode.

Vertex AI

const gemini = new GeminiClient({
  vertexai: true,
  project: 'my-gcp-project',
  location: 'us-central1',
  googleAuthOptions: {
    credentials: trustedServiceAccount,
  },
})

Although project and location are optional in the TypeScript option shape, the Anvia constructor requires non-empty values when it creates the Google client. They may be omitted only when an initialized client is supplied.

Inject a Google client

import { GoogleGenAI } from '@google/genai'

const native = new GoogleGenAI({ apiKey })
const gemini = new GeminiClient({ client: native })

Injection is useful when application code also calls native files, caches, or live APIs.

Embedding configuration

const model = gemini.embeddingModel('gemini-embedding-001', {
  dimensions: 768,
  maxBatchSize: 50,
  taskType: 'RETRIEVAL_DOCUMENT',
  title: 'Product documentation',
})

Changing dimensions or semantic task changes index compatibility. Store those settings alongside collection metadata.

Additional parameters

Completion, images, and transcription accept provider-specific additional parameters through their Anvia request types. Image requests merge a nested config object with the adapter’s response modality and aspect-ratio defaults. Test overrides carefully because the provider can accept combinations that no longer yield an image.

Runtime

The package is ESM, uses Node binary utilities for media conversion, includes declarations, and peers on @anvia/core >=0.7.1 <1.0.0. Validate media and streaming behavior before targeting a non-Node runtime.