lex-gemini

Google Gemini API integration for LegionIO. Generate content, create embeddings, manage files, count tokens, and cache content using Google's Gemini models.

Wraps the Google Gemini REST API as named runners consumable by any LegionIO task chain. Includes resource-specific operations (file management, content caching) not available through the legion-llm unified interface. Use this extension when you need direct access to the full Gemini API surface within the LEX runner/actor lifecycle.

gem install lex-gemini

Or add to your Gemfile:

gem 'lex-gemini'

• generate - Generate text content using Gemini models
• stream_generate - Stream generated content

• embed - Generate embedding vectors from text
• batch_embed - Generate multiple embeddings in one request

• list - List available Gemini models
• get - Get details about a specific model

• count - Count tokens in content

• upload - Upload a file for use with Gemini
• list - List uploaded files
• get - Get file metadata
• delete - Delete an uploaded file

• create - Create cached content for repeated use
• list - List cached contents
• get - Get cached content details
• update - Update cached content expiration
• delete - Delete cached content

Set your API key in your LegionIO settings:

{
  "gemini": {
    "api_key": "AIza..."
  }
}

require 'legion/extensions/gemini/helpers/client'

client = Legion::Extensions::Gemini::Helpers::Client.new(
  api_key: ENV['GEMINI_API_KEY'],
  model: 'gemini-2.0-flash'
)

# Generate content
result = client.generate_content(
  contents: [{ parts: [{ text: 'Explain quantum entanglement in one sentence.' }] }]
)
puts result.dig('candidates', 0, 'content', 'parts', 0, 'text')

# Generate embeddings
embedding_client = Legion::Extensions::Gemini::Helpers::Client.new(
  api_key: ENV['GEMINI_API_KEY'],
  model: 'gemini-embedding-exp'
)
embedding = embedding_client.embed_content(
  content: { parts: [{ text: 'Hello world' }] }
)
puts embedding['embedding']['values'].length

# Count tokens
tokens = client.count_tokens(
  contents: [{ parts: [{ text: 'How many tokens?' }] }]
)
puts tokens['totalTokens']

• faraday >= 2.0 - HTTP client
• faraday-multipart (optional) - required only for multipart file uploads; raw binary upload used as fallback

• Ruby >= 3.4
• LegionIO framework (optional for standalone usage)
• Google Gemini API key (Get one here)

• legion-llm — High-level LLM interface including Gemini via ruby_llm
• extensions-ai/CLAUDE.md — Architecture patterns shared across all AI extensions

0.1.5

MIT