Features

  • Gemini models - Access to Gemini Pro and Gemini Pro Vision
  • Multimodal support - Process text and images
  • Embedding models - Native embedding support
  • Safety settings - Configurable content filtering

Installation

elizaos plugins add @elizaos/plugin-google-genai

Configuration

Environment Variables

# Required
GOOGLE_GENERATIVE_AI_API_KEY=...

# Optional model configuration
# You can use any available Google Gemini model
GOOGLE_SMALL_MODEL=gemini-2.0-flash-001           # Default: gemini-2.0-flash-001
GOOGLE_LARGE_MODEL=gemini-2.5-pro-preview-03-25   # Default: gemini-2.5-pro-preview-03-25
GOOGLE_IMAGE_MODEL=gemini-1.5-flash               # For vision tasks
GOOGLE_EMBEDDING_MODEL=text-embedding-004         # Default: text-embedding-004

# Examples of other available models:
# GOOGLE_SMALL_MODEL=gemini-1.5-flash
# GOOGLE_LARGE_MODEL=gemini-1.5-pro
# GOOGLE_LARGE_MODEL=gemini-pro
# GOOGLE_EMBEDDING_MODEL=embedding-001

Character Configuration

{
  "name": "MyAgent",
  "plugins": ["@elizaos/plugin-google-genai"]
}

Supported Operations

OperationModelsNotes
TEXT_GENERATIONgemini-pro, gemini-pro-visionMultimodal capable
EMBEDDINGembedding-001768-dimensional
OBJECT_GENERATIONAll Gemini modelsStructured output

Model Configuration

The plugin uses three model categories:

  • SMALL_MODEL: Fast, efficient for simple tasks
  • LARGE_MODEL: Best quality, complex reasoning
  • IMAGE_MODEL: Multimodal capabilities (text + vision)
  • EMBEDDING_MODEL: Vector embeddings

You can configure any available Gemini model:

  • Gemini 2.0 Flash (latest)
  • Gemini 2.5 Pro Preview
  • Gemini 1.5 Pro/Flash
  • Gemini Pro (legacy)
  • Any new models Google releases

Safety Configuration

Control content filtering:

// In character settings
{
  "settings": {
    "google_safety": {
      "harassment": "BLOCK_NONE",
      "hate_speech": "BLOCK_MEDIUM_AND_ABOVE",
      "sexually_explicit": "BLOCK_MEDIUM_AND_ABOVE",
      "dangerous_content": "BLOCK_MEDIUM_AND_ABOVE"
    }
  }
}

Usage Tips

  1. Multimodal - Leverage image understanding capabilities
  2. Long Context - Gemini 1.5 Pro supports up to 1M tokens
  3. Rate Limits - Free tier has generous limits

Cost Structure

  • Free tier: 60 queries per minute
  • Paid tier: Higher limits and priority access
  • Embedding calls are separate from generation

External Resources