`@anvia/gemini`

Gemini’s provider adapter connects Anvia to the Gemini Developer API or Vertex AI. It includes completion, embeddings, native Gemini image generation, Imagen generation, transcription, and model listing.


Support	First-party
Version	`0.4.1`
Runtime	ESM, server-side JavaScript
Peer	`@anvia/core >=0.7.1 <1.0.0`

Install

bash

pnpm add @anvia/gemini @anvia/core

Create a Gemini agent

import { AgentBuilder } from '@anvia/core'
import { GeminiClient } from '@anvia/gemini'

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

const agent = new AgentBuilder(
  'assistant',
  gemini.completionModel('gemini-2.5-flash'),
).build()

const result = await agent.prompt('Describe this system in three bullets.').send()
console.log(result.output)

Capabilities

Capability	Factory	Default
Streaming completion	`completionModel()`	`gemini-2.5-flash`
Dense embeddings	`embeddingModel()`	`gemini-embedding-001`
Gemini-native images	`imageGenerationModel()`	`gemini-2.5-flash-image`
Imagen images	`imagenGenerationModel()`	`imagen-4.0-generate-001`
Audio transcription	`transcriptionModel()`	`gemini-2.5-flash`
Model inventory	`listModels()`	Provider model list

Gemini-native images run through generateContent; Imagen uses generateImages. They are separate factories because the provider APIs and supported model families differ. Audio generation is not implemented.

Common patterns

Use Vertex AI

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

const model = vertex.completionModel('gemini-2.5-flash')

Vertex mode uses Google Application Default Credentials unless googleAuthOptions supplies another trusted configuration.

Tune embeddings for retrieval

const queryEmbeddings = gemini.embeddingModel('gemini-embedding-001', {
  taskType: 'RETRIEVAL_QUERY',
  dimensions: 768,
  maxBatchSize: 32,
})

Use matching dimensions and a compatible task configuration when indexing and querying the same vector collection. title is intended for document-oriented embedding tasks.

Compatibility

@anvia/gemini is ESM and uses @google/genai. The client options are a discriminated union: API-key mode cannot include Vertex project settings, while vertexai: true cannot include apiKey. A preconfigured GoogleGenAI client can be injected in either mode.

@anvia/gemini ​

Install ​

Create a Gemini agent ​

Capabilities ​

Common patterns ​

Use Vertex AI ​

Tune embeddings for retrieval ​

Compatibility ​

Continue ​