Get started

Install the adapter with Core:

pnpm add @anvia/core @anvia/gemini

Use an API key for the Gemini Developer API:

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

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

const agent = new Agent({
  id: 'assistant',
  model: gemini.completionModel({
      modelId: 'gemini-3.6-flash'
  }),
})

const result = await agent.generate({
    prompt: 'Summarize this document.'
})

if (result.type === 'response') {
  console.log(result.output)
}

Add embeddings

const documents = gemini.embeddingModel({
    modelId: 'gemini-embedding-001',
    taskType: 'RETRIEVAL_DOCUMENT',
    dimensions: 768
})

const vectors = await documents.embedTexts([
  'Anvia is a provider-neutral TypeScript runtime.',
])

Create a second model configured with RETRIEVAL_QUERY for queries while keeping model and dimensions aligned with indexed documents.

Use Vertex AI

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

Vertex mode and API-key mode are mutually exclusive. See Vertex AI.

Before production

• Keep credentials server-side.
• Use explicit model IDs for each capability.
• Confirm model availability in the selected API or Vertex region.
• Test image, audio, tools, and schemas with the exact model.
• Reindex vectors when dimensions or task configuration changes.
• Treat malformed provider output as an error.