GenAI

AI integration module for Soul

GenAI

The GenAI module provides AI integration capabilities for Soul, supporting multiple AI providers including OpenAI, Anthropic, and Google Gemini. It enables easy AI-powered automation with conversation management, prompt templating, and advanced features like rate limiting and fallback handling.

Basic Usage

Simple AI Chat

Create and use AI with basic configuration:

ai = GenAI.chat("openai")
    .model("gpt-3.5-turbo")
    .configure({
        api_key: "your-api-key",
        max_tokens: 150
    })

response = ai.invoke("Hello, how are you?")
println(response.text)

Multi-Provider Setup

Configure different AI providers:

// OpenAI setup
openai = GenAI.chat("openai")
    .model("gpt-4")
    .configure({
        api_key: "your-openai-key",
        max_tokens: 200
    })

// Anthropic setup
claude = GenAI.chat("anthropic")
    .model("claude-3-haiku")
    .configure({
        api_key: "your-anthropic-key",
        max_tokens: 300
    })

// Google Gemini setup
gemini = GenAI.chat("google")
    .model("gemini-pro")
    .configure({
        api_key: "your-google-key",
        max_tokens: 250
    })

Conversation Management

Basic Conversation

Create and manage multi-turn conversations:

conversation = GenAI.createConversation()

// Add messages to conversation
conversation.addSystemMessage("You are a helpful assistant.")
conversation.addUserMessage("What is the capital of France?")

// Get AI response with conversation context
ai = GenAI.chat("openai").model("gpt-3.5-turbo")
response = ai.invokeConversation(conversation)

// Add AI response to conversation
conversation.addAssistantMessage(response.text)

// Continue conversation
conversation.addUserMessage("What about Germany?")
response = ai.invokeConversation(conversation)

Conversation Utilities

Use conversation helper methods:

conversation = GenAI.createConversation()

// Add various message types
conversation.addSystemMessage("You are a coding assistant.")
conversation.addUserMessage("How do I create a function in Soul?")

// Get conversation info
messageCount = conversation.length()
messages = conversation.getMessages()
lastUserMessage = conversation.getLastUserMessage()

// Summarize conversation
summary = conversation.summarize()
println(summary)  // "Conversation with 1 user messages and 0 assistant messages"

Prompt Templates

Creating Templates

Build reusable prompt templates:

// Create a template with variables
template = GenAI.createTemplate("Hello {name}, welcome to {platform}!")

// Render template with variables
rendered = template.render({
    name: "Alice",
    platform: "Soul"
})
println(rendered)  // "Hello Alice, welcome to Soul!"

// Validate template variables
validation = template.validate({
    name: "Alice"
    // Missing 'platform' variable
})
println(validation)  // "Missing variables: platform"

Registered Templates

Register and reuse templates globally:

// Register a template
GenAI.registerTemplate("greeting", "Hello {name}! How can I help you with {topic}?")

// Use registered template
template = GenAI.getTemplate("greeting")
message = template.render({
    name: "Bob",
    topic: "programming"
})

ai = GenAI.chat("openai").model("gpt-3.5-turbo")
response = ai.invoke(message)

Advanced Features

Token Counting

Estimate token usage:

text = "This is a sample text for token counting"
model = "gpt-3.5-turbo"

tokenCount = GenAI.countTokens(text, model)
println("Estimated tokens: " + tokenCount)

Text Preprocessing

Clean and prepare text for AI:

rawText = "  This    has    extra   spaces  \n\n  and   newlines  "
cleanedText = GenAI.preprocessText(rawText)
println(cleanedText)  // "This has extra spaces and newlines"

Response Validation

Validate AI responses:

response = ai.invoke("What is 2+2?")
isValid = GenAI.validateResponse(response)

if (isValid) {
    println("Response is valid: " + response.text)
} else {
    println("Response is invalid or contains errors")
}

Model Capabilities

Get information about AI models:

capabilities = GenAI.getModelCapabilities("gpt-4")
println("Max tokens: " + capabilities.max_tokens)
println("Context window: " + capabilities.context_window)
println("Supports functions: " + capabilities.supports_functions)
println("Supports vision: " + capabilities.supports_vision)

Batch Processing

Creating Batches

Process multiple prompts efficiently:

prompts = [
    "What is the capital of France?",
    "What is the capital of Germany?",
    "What is the capital of Italy?"
]

batch = GenAI.createBatch(prompts)
println("Batch ID: " + batch.getId())
println("Estimated tokens: " + batch.getEstimatedTokens())

Rate Limiting

Basic Rate Limiting

Implement rate limiting for API calls:

// Create rate limiter: 100 requests per minute, max 10 concurrent
rateLimiter = GenAI.createRateLimiter(100, 10)

// Check if request is allowed
if (rateLimiter.checkLimit()) {
    // Make API call
    response = ai.invoke("Hello")
    rateLimiter.recordRequest()
    
    println("Remaining requests: " + rateLimiter.getRemainingRequests())
    println("Current usage: " + rateLimiter.getCurrentUsage())
} else {
    println("Rate limit exceeded")
}

Retry Handling

Retry Configuration

Handle API failures with retry logic:

// Create retry handler with exponential backoff
retryConfig = {
    max_retries: 3,
    base_delay: 1000,  // 1 second
    exponential_backoff: true
}

retryHandler = GenAI.createRetryHandler(retryConfig)

// Get delay for retry attempt
attempt = 2
delay = retryHandler.getDelay(attempt)
println("Retry delay: " + delay + "ms")

Fallback Management

Model Fallbacks

Set up fallback models for reliability:

// Create fallback chain
fallbackModels = ["gpt-4", "gpt-3.5-turbo", "claude-3-haiku"]
fallbackManager = GenAI.createFallbackManager(fallbackModels)

// Get current model
currentModel = fallbackManager.getCurrentModel()
println("Using model: " + currentModel)

// Switch to next model on failure
nextModel = fallbackManager.getNextModel()
if (nextModel != null) {
    println("Switching to: " + nextModel)
}

Complete Example

AI-Powered Chat Assistant

soul genesis() {
    // Configure AI
    ai = GenAI.chat("openai")
        .model("gpt-3.5-turbo")
        .configure({
            api_key: "your-api-key",
            max_tokens: 150
        })
    
    // Create conversation
    conversation = GenAI.createConversation()
    conversation.addSystemMessage("You are a helpful programming assistant.")
    
    // Create rate limiter
    rateLimiter = GenAI.createRateLimiter(60, 5)  // 60 requests per minute
    
    // Create template for user queries
    GenAI.registerTemplate("code_help", "Help me with {language}: {question}")
    
    // Simulate user interactions
    userQueries = [
        {language: "Soul", question: "How do I create a function?"},
        {language: "Soul", question: "What is the syntax for classes?"},
        {language: "Soul", question: "How do I handle errors?"}
    ]
    
    template = GenAI.getTemplate("code_help")
    
    for (query in userQueries) {
        // Check rate limit
        if (!rateLimiter.checkLimit()) {
            println("Rate limit exceeded, waiting...")
            continue
        }
        
        // Render prompt from template
        prompt = template.render(query)
        
        // Add to conversation
        conversation.addUserMessage(prompt)
        
        // Get AI response
        response = ai.invokeConversation(conversation)
        
        // Record request
        rateLimiter.recordRequest()
        
        // Validate response
        if (GenAI.validateResponse(response)) {
            println("Question: " + query.question)
            println("Answer: " + response.text)
            println("Tokens used: " + response.usage.total_tokens)
            println("---")
            
            // Add response to conversation
            conversation.addAssistantMessage(response.text)
        } else {
            println("Invalid response received")
        }
    }
    
    // Print conversation summary
    println("Final conversation summary:")
    println(conversation.summarize())
}

Configuration Options

Provider-Specific Settings

Different providers support different configuration options:

// OpenAI configuration
openaiConfig = {
    api_key: "your-openai-key",
    max_tokens: 150,
    temperature: 0.7,
    top_p: 1.0,
    frequency_penalty: 0.0,
    presence_penalty: 0.0
}

// Anthropic configuration
anthropicConfig = {
    api_key: "your-anthropic-key",
    max_tokens: 200,
    temperature: 0.5
}

// Google configuration
googleConfig = {
    api_key: "your-google-key",
    max_tokens: 250,
    temperature: 0.8
}

Error Handling

Robust Error Handling

Handle AI API errors gracefully:

try {
    ai = GenAI.chat("openai")
        .model("gpt-3.5-turbo")
        .configure({
            api_key: "invalid-key"
        })
    
    response = ai.invoke("Hello")
    println(response.text)
    
} catch (error) {
    println("AI API error: " + error.message)
    
    // Try fallback model
    fallbackAI = GenAI.chat("anthropic")
        .model("claude-3-haiku")
        .configure({
            api_key: "backup-key"
        })
    
    try {
        response = fallbackAI.invoke("Hello")
        println("Fallback response: " + response.text)
    } catch (fallbackError) {
        println("Fallback also failed: " + fallbackError.message)
    }
}

Best Practices

1. Use rate limiting: Prevent API quota exhaustion
2. Implement fallbacks: Have backup models ready
3. Validate responses: Check for errors before processing
4. Manage conversations: Track context for multi-turn chats
5. Monitor token usage: Keep track of API costs

// Good - comprehensive AI setup
soul createRobustAI() {
    // Create AI with fallback
    fallbackModels = ["gpt-4", "gpt-3.5-turbo", "claude-3-haiku"]
    fallbackManager = GenAI.createFallbackManager(fallbackModels)
    
    // Rate limiting
    rateLimiter = GenAI.createRateLimiter(100, 10)
    
    // Retry handling
    retryHandler = GenAI.createRetryHandler({
        max_retries: 3,
        base_delay: 1000,
        exponential_backoff: true
    })
    
    // Main AI instance
    ai = GenAI.chat("openai")
        .model(fallbackManager.getCurrentModel())
        .configure({
            api_key: "your-api-key",
            max_tokens: 200
        })
    
    return {
        ai: ai,
        rateLimiter: rateLimiter,
        retryHandler: retryHandler,
        fallbackManager: fallbackManager
    }
}

The GenAI module provides a comprehensive AI integration platform for Soul, enabling sophisticated AI-powered applications with robust error handling, conversation management, and multi-provider support.