Rbai is a lightweight Ruby client that provides a unified interface to major LLM/AI vendors: Google Gemini, OpenAI GPT, and Anthropic Claude. It abstracts over different REST APIs and response formats, making it easier to send prompts and retrieve text responses with minimal setup.

• Support for Google Generative Language API (Gemini)
• Support for OpenAI Chat Completions API
• Support for Anthropic Claude Messages API
• Unified generate_content method for prompts
• Automatic response parsing and text extraction
• Configurable system instructions and generation parameters

Add this line to your application's Gemfile:

gem "rbai", github: "tlkahn/rbai"

or:

gem "rbai", git: "https://github.com/tlkahn/rbai"

And then execute:

bundle install

Or install manually:

gem install rbai

Export your API keys as environment variables:

# Google Gemini
export GOOGLE_API_KEY="your-google-api-key"

# or alternative
export GOOGLE_GENAI_API_KEY="your-google-genai-api-key"

# OpenAI
export OPENAI_API_KEY="your-openai-api-key"

# Anthropic Claude
export CLAUDE_API_KEY="your-claude-api-key"

# or alternative
export ANTHROPIC_API_KEY="your-anthropic-api-key"

require "rbai"

# Initialize client
client = Rbai::Client.new(provider: :openai)

# Send a prompt
response = client.generate_content("Write a haiku about Ruby programming.")
puts response

client = Rbai::Client.new(provider: :claude)

response = client.generate_content(
  "Explain quantum entanglement simply.",
  system_instruction: "You are a physics tutor who uses plain language."
)

puts response

client = Rbai::Client.new(provider: :google)

response = client.generate_content(
  "List three benefits of functional programming.",
  generation_config: { temperature: 0.7, maxOutputTokens: 200 }
)

puts response

Initialize a client for a given provider. If api_key is not supplied, it will be read from environment variables.

• provider — one of :google, :openai, :claude
• api_key (optional) — explicit API key

Send a prompt to the selected provider and return the response text.

• prompt — user input (string)
• system_instruction — optional system role content
• generation_config — provider-specific config (hash)
• model_id — override default model (string)

Returns: plain text string response.

If the HTTP request fails or the provider returns an error, an exception is raised:

begin
  response = client.generate_content("Hello world")
rescue => e
  warn "Error: #{e.message}"
end

To run tests and lint:

bundle exec rake test
bundle exec rubocop

MIT