openai-conversation-manager

OpenAI Prompt Script

This script allows you to interact with OpenAI's API to generate responses based on predefined or custom prompts. It supports three modes of operation: JSON Config Mode, File Input Mode, and Direct Input Mode. You can also simulate recurring conversations by including a history of previous exchanges within each request.

Table of Contents

• Overview
• Prerequisites
• Usage Modes
  • JSON Config Mode
  • File Input Mode
  • Direct Input Mode
• Parameters
• Recurring Conversations
• Examples
• Error Handling
• Setup
• Configuration Example
• License

---

Overview

This script sends prompts to the OpenAI API using various input modes, allowing for flexible, customizable interactions. It’s designed to:

1. Load agent prompts and settings from a JSON configuration file.
2. Use direct file paths or input strings for both agent and user prompts.
3. Retain context for ongoing conversations by including previous exchanges in each request.

Prerequisites

• Node.js and npm installed.
• An OpenAI API key stored in a .env file (for security).

.env File Example

OPENAI_API_KEY=your_openai_api_key_here
CONFIG_PATH=path_to_config_file

Conversation Manager Documentation

Overview

The Conversation Manager is a comprehensive tool for interacting with the OpenAI API in a way that maintains context across conversations, handles logging, and manages system prompts. It enables developers to create, maintain, and retrieve conversations for users with ease.

Usage Guide

Initializing a Conversation Manager

To start a conversation, initialize the ConversationManager class. If you're not saving any user data, you can instantiate it without parameters.

// Create a new conversation without saving user data
const conversation = new ConversationManager();

// Create a conversation for a specific user
const conversation = new ConversationManager('user123');

This loads the user data, including past conversations, if any.

Setting Up the System

By default, the ConversationManager uses predefined system settings. There are three modes for changing this setup:

1. Direct Mode: Directly set the prompt and options.
2. File Mode: Load the system settings from a file.
3. Config Mode: Use settings from a configuration file.

// Setting the system with direct input
conversation.setSystem('direct', {
  agentPrompt: 'You are a helpful assistant.',
  model: 'gpt-3.5-turbo',
  temperature: 0.7,
  conversationMaxTokens: 1000,
  responseTokens: 150,
});

// Setting the system from a file
conversation.setSystem('file', { agentFilePath: 'path/to/agentFile.txt' });

// Setting the system using config mode
conversation.setSystem('config', { modelId: 'empathy_coach' });

Adding User Messages

Use the addMessage method to add a user message to the conversation.

conversation.addMessage('Can you provide me with some tips for productivity?');

Sending a Message to the API

The callAPI method sends the current conversation history to OpenAI and appends the assistant's response.

const response = await conversation.callAPI();
console.log('Assistant response:', response);

Saving and Loading Conversation History

The saveHistory method saves the current conversation, while loadLatestConversation loads the most recent conversation for a user.

conversation.saveHistory(); // Save the current conversation

conversation.loadLatestConversation(); // Load the last conversation for the user

Retrieving Conversation History

You can use the getHistory method to retrieve the current conversation history, with an option to exclude system messages.

const history = conversation.getHistory(true); // Retrieve history without system messages
console.log(history);

Method: loadConversationById

Usage: Loads a specific conversation by its unique conversationId for a user, setting it as the active conversation.

Purpose: Enables resuming or continuing a specific previous conversation, maintaining context and continuity.

Example:

conversation.loadConversationById("2"); // Loads conversation with ID "2" for the user

Additional Methods

• startNewConversation(): Begins a new conversation for the user.
• deleteHistory(): Deletes the current conversation history but retains system messages.

Example Scripts

Simple Interaction Script

import { ConversationManager } from 'path/to/ConversationManager.js';

(async () => {
  const conversation = new ConversationManager('user123');
  conversation.setSystem('config', { modelId: 'empathy_coach' });
  conversation.addMessage('How do I improve my communication skills?');
  const response = await conversation.callAPI();
  console.log('Response:', response);
  conversation.saveHistory();
})();

Error Handling

The ConversationManager includes built-in error handling. Ensure that file paths are valid and the modelId matches an entry in your configuration.

Setup

1. Clone the Repository:

   git clone your-repo-url
   cd your-repo-name

2. Install Dependencies:

   npm install dotenv node-fetch

3. Create .env File:

   • Add your OpenAI API key in the .env file as shown in Prerequisites. You may copy the template, env.example to .env to do so.

4. Create config.json:

   • Set up your JSON configuration file as described below.

Configuration Example

The config.json file contains predefined agent configurations with unique IDs, default settings, and file paths for agent prompts.

{
  "defaults": {
    "model": "gpt-4o-mini",
    "conversationMaxTokens": 3000,
    "responseTokens": 200,
    "temperature": 0.7,
    "logPath": "logs"
  },
  "models": {
    "poet": {
      "agent_file": "agents/poet.txt",
      "conversationMaxTokens": 300,
      "responseTokens": 75,
      "temperature": 0.6,
      "model": "gpt-4o-mini"
    },
    "general_question_confirmer": {
      "agent_file": "agents/general_question_confirmer.txt",
      "conversationMaxTokens": 3000,
      "responseTokens": 200,
      "temperature": 0.6,
      "model": "gpt-4o-mini"
    }
  }
}

Each models entry in the configuration should include:

• id: A unique identifier for the agent configuration.
• agent_file: Path to the file containing the agent’s prompt.
• max_tokens: Default maximum token count for responses.
• temperature: Default randomness control for responses.
• model: OpenAI model name.

User Data Structure

The ConversationManager stores user data in a structured JSON format for easy retrieval and persistence. Each user has a dedicated file named username.json that includes metadata and all their conversation histories. Below is the structure of the user data:

File Structure

Each user's data is stored in a single JSON file named after their userId, with the following structure:

{
  "userId": "username",
  "totalConversations": 3,
  "conversations": [
    {
      "conversationId": "1",
      "name": "How can I stay focused?",
      "timestamp": "2024-11-03T10:00:00Z",
      "messages": [
        { "role": "system", "content": "Welcome to the session." },
        { "role": "user", "content": "How can I stay focused?" },
        { "role": "assistant", "content": "Set clear goals, eliminate distractions, and take breaks." }
      ]
    },
    {
      "conversationId": "2",
      "name": "Productivity tips",
      "timestamp": "2024-11-04T14:30:00Z",
      "messages": [
        { "role": "system", "content": "Productivity coaching session started." },
        { "role": "user", "content": "Give me a quick productivity tip." },
        { "role": "assistant", "content": "Try the two-minute rule for small tasks." }
      ]
    },
    {
      "conversationId": "3",
      "name": "Empathy session",
      "timestamp": "2024-11-05T09:45:00Z",
      "messages": [
        { "role": "system", "content": "Empathy coaching session initiated." },
        { "role": "user", "content": "I need some empathy. Can you help me understand my feelings?" },
        { "role": "assistant", "content": "Of course, I'm here for you. What are you feeling at the moment?" }
      ]
    }
  ]
}

• userId: The unique identifier for the user.
• totalConversations: The total number of conversations stored for the user.
• conversations: An array of conversations, each containing:
  • conversationId: The unique ID of the conversation.
  • name: A brief title derived from the first user message or system prompt.
  • timestamp: The timestamp of when the conversation was created.
  • messages: An array of message objects, each containing:
    • role: The sender's role (system, user, assistant).
    • content: The content of the message.

License

This project is licensed under the MIT License.