@turbot/guardrails-lib-ai

Guardrails Lib AI

A unified interface for multiple AI language model providers. This wrapper provides a consistent API to interact with various AI providers while handling provider-specific configurations and fallback mechanisms.

Features

• Unified Interface: Single API to interact with multiple AI providers
• Provider Support:
  • OpenAI (GPT-4, GPT-3.5)
  • Anthropic (Claude)
• Error Handling: Robust error handling with Turbot error types
• Logging: Structured logging with Turbot logging utilities
• Type Safety: Full TypeScript support
• Proxy Support: Built-in support for HTTP/HTTPS proxies

Installation

npm install guardrails-lib-ai

Usage

const AI = require('guardrails-lib-ai');

// Initialize with OpenAI using direct API key and proxy
const ai = new AI({
  provider: 'openai',
  modelName: 'gpt-4',
  apiKey: 'your-openai-api-key',  // Direct API key
  system: 'You are a helpful AI assistant.',
  proxyUrl: 'http://your-proxy-server:port'  // Optional proxy configuration
});

// Query the model
try {
  const response = await ai.generate('Explain quantum computing in simple terms');
  console.log(response);
} catch (error) {
  // Errors are instances of Turbot error types
  if (error.name === 'ProviderError') {
    console.error('Provider error:', error.message);
  } else if (error.name === 'BadConfigurationError') {
    console.error('Configuration error:', error.message);
    console.error('Missing parameters:', error.missingParams);
  } else {
    console.error('Unexpected error:', error.message);
  }
}

// Initialize with Anthropic
const claude = new AI({
  provider: 'anthropic',
  modelName: 'claude-3-opus-20240229',
  apiKey: 'your-anthropic-api-key',  // Direct API key
  system: 'You are a helpful AI assistant.'
});

// Query with streaming
try {
  const response = await claude.generate('Write a short poem about AI');
  console.log(response);
} catch (error) {
  console.error('Error:', error);
}

Proxy Configuration

The library supports proxy configuration in two ways:

1. Direct Configuration:

   const ai = new AI({
     provider: 'openai',
     modelName: 'gpt-4',
     proxyUrl: 'http://your-proxy-server:port'
   });

2. No Proxy:

   const ai = new AI({
     provider: 'openai',
     modelName: 'gpt-4'
     // No proxy configuration
   });

The proxy configuration is supported for both OpenAI and Anthropic providers. When using a proxy:

• All API requests will be routed through the specified proxy server
• The proxy configuration is applied consistently across all providers
• Both HTTP and HTTPS proxies are supported

Error Handling

The wrapper uses Turbot's error handling system (@turbot/errors) to provide structured error types:

Error Types

• BadConfigurationError: Thrown when required configuration is missing

  try {
    const ai = new AI({}); // Missing required config
  } catch (error) {
    if (error.name === 'BadConfigurationError') {
      console.error('Missing parameters:', error.missingParams);
    }
  }

• ProviderError: Thrown for provider-specific issues

  try {
    await ai.generate('prompt');
  } catch (error) {
    if (error.name === 'ProviderError') {
      console.error('Provider error:', error.message);
      console.error('Provider:', error.provider);
    }
  }

• ModelError: Thrown for model-specific issues

  try {
    await ai.generate('prompt');
  } catch (error) {
    if (error.name === 'ModelError') {
      console.error('Model error:', error.message);
      console.error('Model:', error.modelName);
    }
  }

Configuration Options

The AI constructor accepts the following configuration options:

{
  provider: string;      // The AI provider to use ('openai' or 'anthropic')
  modelName: string;     // The specific model to use (e.g., 'gpt-4', 'claude-3-opus-20240229')
  apiKey: string;        // API key (required)
  system?: string;       // Optional system prompt
  proxyUrl?: string;     // Optional proxy URL
  max_tokens?: number;   // Optional max tokens
  temperature?: number;  // Optional temperature
}

API Key Configuration

You must provide API keys directly in the configuration:

const ai = new AI({
  provider: 'openai',
  modelName: 'gpt-4',
  apiKey: 'your-api-key-here'  // Direct API key (required)
});

Supported Providers and Models

• OpenAI
• Anthropic

Node.js Version

Requires Node.js >= 18.0.0