@charivo/llm-client-openai

OpenAI ChatGPT client for Charivo LLM system.

Features

• 🤖 OpenAI Integration - Works with GPT-4, GPT-3.5, and other OpenAI models
• 💬 Streaming Support - Optional streaming responses
• 🎯 Type-Safe - Full TypeScript support
• 🔧 Configurable - Customize model, temperature, and other parameters

Installation

pnpm add @charivo/llm-client-openai @charivo/core

Usage

Basic Setup

import { OpenAILLMClient } from "@charivo/llm-client-openai";

const client = new OpenAILLMClient({
  apiKey: "your-openai-api-key",
  model: "gpt-4"
});

const response = await client.call([
  { role: "user", content: "Hello!" }
]);

console.log(response); // "Hello! How can I help you today?"

With LLMManager (Recommended)

import { OpenAILLMClient } from "@charivo/llm-client-openai";
import { createLLMManager } from "@charivo/llm-core";

const client = new OpenAILLMClient({
  apiKey: "your-openai-api-key",
  model: "gpt-4-turbo-preview"
});

const llmManager = createLLMManager(client);

llmManager.setCharacter({
  id: "assistant",
  name: "Hiyori",
  personality: "Cheerful and helpful"
});

const response = await llmManager.generateResponse({
  id: "1",
  content: "Hello!",
  timestamp: new Date(),
  type: "user"
});

Custom Configuration

const client = new OpenAILLMClient({
  apiKey: "your-openai-api-key",
  model: "gpt-4",
  temperature: 0.7,      // Creativity (0.0 - 2.0)
  maxTokens: 1000,       // Max response length
  topP: 0.9,             // Nucleus sampling
  frequencyPenalty: 0.5, // Reduce repetition
  presencePenalty: 0.5   // Encourage new topics
});

API Reference

Constructor

new OpenAILLMClient(config: OpenAILLMClientConfig)

Config Options:

• apiKey: string - Your OpenAI API key (required)
• model?: string - Model name (default: "gpt-4")
• temperature?: number - Sampling temperature 0-2 (default: 0.7)
• maxTokens?: number - Max tokens in response (default: 1000)
• topP?: number - Nucleus sampling 0-1 (default: 1.0)
• frequencyPenalty?: number - Frequency penalty 0-2 (default: 0)
• presencePenalty?: number - Presence penalty 0-2 (default: 0)

Methods

call(messages)

Send messages and get a response.

const response = await client.call([
  { role: "user", content: "Hi" },
  { role: "assistant", content: "Hello!" },
  { role: "user", content: "How are you?" }
]);

Note: This client wraps the OpenAI provider and is suitable for development/testing. For production, use @charivo/llm-client-remote with a server-side provider to keep API keys secure.

Supported Models

GPT-4 Models

• gpt-4 - Most capable, best for complex tasks
• gpt-4-turbo-preview - Faster, cheaper, 128K context
• gpt-4-0125-preview - Latest GPT-4 Turbo
• gpt-4-1106-preview - Previous GPT-4 Turbo

GPT-3.5 Models

• gpt-3.5-turbo - Fast and cost-effective
• gpt-3.5-turbo-16k - Extended context window

Configuration Guide

Temperature

Controls randomness (0.0 - 2.0):

• 0.0-0.3: Focused, deterministic (good for facts)
• 0.4-0.7: Balanced (good for conversation)
• 0.8-1.0: Creative (good for storytelling)
• 1.1-2.0: Very creative (experimental)

// Factual assistant
const factualClient = new OpenAILLMClient({
  apiKey: "...",
  temperature: 0.2
});

// Creative storyteller
const creativeClient = new OpenAILLMClient({
  apiKey: "...",
  temperature: 0.9
});

Max Tokens

Limits response length:

• Short responses: 100-300
• Normal conversation: 500-1000
• Long-form content: 1500-2000

const client = new OpenAILLMClient({
  apiKey: "...",
  maxTokens: 500 // Concise responses
});

Penalties

Reduce repetition and encourage diversity:

const client = new OpenAILLMClient({
  apiKey: "...",
  frequencyPenalty: 0.5, // Reduce word repetition
  presencePenalty: 0.5   // Encourage new topics
});

Error Handling

try {
  const response = await client.call(messages);
} catch (error) {
  if (error.code === "insufficient_quota") {
    console.error("OpenAI quota exceeded");
  } else if (error.code === "invalid_api_key") {
    console.error("Invalid API key");
  } else {
    console.error("OpenAI error:", error);
  }
}

Environment Variables

Use environment variables for API keys:

# .env
OPENAI_API_KEY=sk-...

const client = new OpenAILLMClient({
  apiKey: process.env.OPENAI_API_KEY!
});

Cost Optimization

1. Use GPT-3.5 for simple tasks: 10x cheaper than GPT-4
2. Limit maxTokens: Reduce response length
3. Cache responses: Store common responses
4. Use temperature wisely: Lower temperature = more deterministic = better caching

// Cost-effective setup
const client = new OpenAILLMClient({
  apiKey: "...",
  model: "gpt-3.5-turbo", // Cheaper
  maxTokens: 300,         // Shorter responses
  temperature: 0.3        // More cacheable
});

License

MIT