LLM Conductor

gem 'llm_conductor'

# Direct prompt generation - easiest way to get started
response = LlmConductor.generate(
  model: 'gpt-5-mini',
  prompt: 'Explain quantum computing in simple terms'
)

puts response.output           # The generated text
puts response.total_tokens     # Token usage
puts response.estimated_cost   # Cost in USD

# Use built-in text summarization template
response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :summarize_text,
  data: {
    text: 'Ekohe (ee-koh-hee) means "boundless possibility." Our way is to make AI practical, achievable, and most importantly, useful for you — and we prove it every day. With almost 16 years of wins under our belt, a market-leading 24-hr design & development cycle, and 5 offices in the most vibrant cities in the world, we surf the seas of innovation. We create efficient, elegant, and scalable digital products — delivering the right interactive solutions to achieve your audience and business goals. We help you transform. We break new ground across the globe — from AI and ML automation that drives the enterprise, to innovative customer experiences and mobile apps for startups. Our special sauce is the care, curiosity, and dedication we offer to solve for your needs. We focus on your success and deliver the most impactful experiences in the most efficient manner. Our clients tell us we partner with them in a trusted and capable way, driving the right design and technical choices.',
    max_length: '20 words',
    style: 'professional and engaging',
    focus_areas: ['core business', 'expertise', 'target market'],
    audience: 'potential investors',
    include_key_points: true,
    output_format: 'paragraph'
  }
)

# Response object provides rich information
if response.success?
  puts "Generated: #{response.output}"
  puts "Tokens: #{response.total_tokens}"
  puts "Cost: $#{response.estimated_cost || 'N/A (free model)'}"
else
  puts "Error: #{response.metadata[:error]}"
end

LlmConductor.configure do |config|
  # Default settings
  config.default_model = 'gpt-5-mini'
  config.default_vendor = :openai
  config.timeout = 30
  config.max_retries = 3
  config.retry_delay = 1.0

  # Provider configurations
  config.openai(
    api_key: ENV['OPENAI_API_KEY'],
    organization: ENV['OPENAI_ORG_ID'] # Optional
  )

  config.anthropic(
    api_key: ENV['ANTHROPIC_API_KEY']
  )

  config.gemini(
    api_key: ENV['GEMINI_API_KEY']
  )

  config.ollama(
    base_url: ENV['OLLAMA_ADDRESS'] || 'http://localhost:11434'
  )
end

response = LlmConductor.generate(
  model: 'gpt-5-mini',  # Auto-detects OpenAI
  prompt: 'Your prompt here'
)

response = LlmConductor.generate(
  model: 'claude-3-5-sonnet-20241022',  # Auto-detects Anthropic
  prompt: 'Your prompt here'
)

# Or explicitly specify vendor
response = LlmConductor.generate(
  model: 'claude-3-5-sonnet-20241022',
  vendor: :anthropic,
  prompt: 'Your prompt here'
)

response = LlmConductor.generate(
  model: 'gemini-2.5-flash',  # Auto-detects Gemini
  prompt: 'Your prompt here'
)

# Or explicitly specify vendor
response = LlmConductor.generate(
  model: 'gemini-2.5-flash',
  vendor: :gemini,
  prompt: 'Your prompt here'
)

response = LlmConductor.generate(
  model: 'llama3.2',  # Auto-detects Ollama for non-GPT/Claude/Gemini models
  prompt: 'Your prompt here'
)

response = LlmConductor.generate(
  model: 'llama3.2',  # Auto-detects Ollama for non-GPT models
  prompt: 'Your prompt here'
)

class CompanyAnalysisPrompt < LlmConductor::Prompts::BasePrompt
  def render
    <<~PROMPT
      Company: #{name}
      Domain: #{domain_name}
      Description: #{truncate_text(description, max_length: 1000)}

      Please analyze this company and provide:
      1. Core business model
      2. Target market 
      3. Competitive advantages
      4. Growth potential

      Format as JSON.
    PROMPT
  end
end

# Register the prompt
LlmConductor::PromptManager.register(:detailed_analysis, CompanyAnalysisPrompt)

# Use the registered prompt
response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :detailed_analysis,
  data: {
    name: 'Ekohe',
    domain_name: 'ekohe.com',
    description: 'A leading AI company...'
  }
)

# Parse structured responses
analysis = response.parse_json
puts analysis

class CompanyDataBuilder < LlmConductor::DataBuilder
  def build
    {
      id: source_object.id,
      name: source_object.name,
      description: format_for_llm(source_object.description, max_length: 500),
      industry: extract_nested_data(:data, 'categories', 'primary'),
      metrics: build_metrics,
      summary: build_company_summary,
      domain_name: source_object.domain_name

    }
  end

  private

  def build_metrics
    {
      employees: format_number(source_object.employee_count),
      revenue: format_number(source_object.annual_revenue),
      growth_rate: "#{source_object.growth_rate}%"
    }
  end

  def build_company_summary
    name = safe_extract(:name, default: 'Company')
    industry = extract_nested_data(:data, 'categories', 'primary')
    "#{name} is a #{industry} company..."
  end
end

# Usage
company = Company.find(123)
data = CompanyDataBuilder.new(company).build

response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :detailed_analysis, 
  data: data
)

response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :featured_links,
  data: {
    htmls: '<html>...</html>',
    current_url: 'https://example.com'
  }
)

response = LlmConductor.generate(
  model: 'gpt-5-mini', 
  type: :summarize_htmls,
  data: { htmls: '<html>...</html>' }
)

response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :summarize_description,
  data: {
    name: 'Company Name',
    description: 'Long description...',
    industries: ['Tech', 'AI']
  }
)

response = LlmConductor.generate(
  model: 'gpt-5-mini',
  type: :custom,
  data: {
    template: "Analyze this data: %{data}",
    data: "Your data here"
  }
)

response = LlmConductor.generate(...)

# Main content
response.output           # Generated text
response.success?         # Boolean success status

# Token information
response.input_tokens     # Input tokens used
response.output_tokens    # Output tokens generated  
response.total_tokens     # Total tokens

# Cost tracking (for supported models)
response.estimated_cost   # Estimated cost in USD

# Metadata
response.model           # Model used
response.metadata        # Hash with vendor, timestamp, etc.

# Structured data parsing
response.parse_json                    # Parse as JSON
response.extract_code_block('json')    # Extract code blocks

response = LlmConductor.generate(
  model: 'gpt-5-mini',
  prompt: 'Your prompt'
)

if response.success?
  puts response.output
else
  puts "Error: #{response.metadata[:error]}"
  puts "Failed model: #{response.model}"
end

# Exception handling for critical errors
begin
  response = LlmConductor.generate(...)
rescue LlmConductor::Error => e
  puts "LLM Conductor error: #{e.message}"
rescue StandardError => e
  puts "General error: #{e.message}" 
end

module LlmConductor
  module Clients
    class CustomClient < BaseClient
      private

      def generate_content(prompt)
        # Implement your provider's API call
        your_custom_api.generate(prompt)
      end
    end
  end
end

module LlmConductor
  module Prompts
    def prompt_custom_analysis(data)
      <<~PROMPT
        Custom analysis for: #{data[:subject]}
        Context: #{data[:context]}
        
        Please provide detailed analysis.
      PROMPT
    end
  end
end