@mencraft/lore-headless v0.1.22

Lore Headless

A headless library for chat streaming functionality that works with React and other frontend frameworks.

Installation

npm install @mencraft/lore-headless

Configuration

When using this library, you'll need to provide the following authentication values:

// Authentication credentials
const auth = {
  clientId: 'your-client-id',     // Your client ID
  token: 'your-firebase-id-token' // Your Firebase ID token
};

// Graph ID for chat operations
const graphId = 'your-graph-id';  // Your graph ID

API URLs

By default, the API client is configured for production URLs. You can specify different modes or use custom URLs:

// Default usage (production mode)
const apiClient = new LoreClient();

// Specify a different mode
const stagingClient = new LoreClient('staging');
const devClient = new LoreClient('development');

// Use custom URLs (mode will be ignored when custom URLs are provided)
const customClient = new LoreClient(
  'production', // This will be ignored when custom URLs are provided
  'https://your-custom-chat-url.com',
  'https://your-custom-validation-url.com'
);

Available modes:

• 'production' (default): Uses production API URLs
• 'staging': Uses staging API URLs
• 'development': Uses local development URLs (localhost)

Custom URLs: When providing custom URLs, both customChatUrl and customValidationUrl must be provided together. If custom URLs are specified, the mode parameter will be ignored.

In a real application, you would typically set the authentication values using environment variables:

// Authentication credentials
const auth = {
  clientId: process.env.CLIENT_ID,
  token: process.env.FIREBASE_TOKEN
};

// Graph ID for chat operations
const graphId = process.env.GRAPH_ID;

Usage

Core API (Framework Agnostic)

import { LoreClient, ChatStreamManager, ChatRequest, ChatStreamCallbacks } from 'lore-headless';

// Create an API client
// Default usage (production mode)
const apiClient = new LoreClient();

// Or specify a different mode
// const apiClient = new LoreClient('staging');
// const apiClient = new LoreClient('development');

// Or use custom URLs
// const apiClient = new LoreClient(
//   'production', // mode will be ignored
//   'https://your-custom-chat-url.com',
//   'https://your-custom-validation-url.com'
// );

// Set up authentication
const auth = {
  clientId: 'your-client-id',
  token: 'your-firebase-id-token'
};

// Define callbacks
const callbacks: ChatStreamCallbacks = {
  onTextStream: (text) => {
    console.log('Received text:', text);
  },
  onBatchData: (data) => {
    console.log('Received batch data:', data);
  },
  onError: (error) => {
    console.error('Error:', error);
  },
  onEnd: () => {
    console.log('Stream ended');
  }
};

// Create a chat stream manager
const chatManager = new ChatStreamManager(apiClient, auth, callbacks);

// Start a chat stream
const request: ChatRequest = {
  user: 'user-id',
  session: 'session-id',
  query: 'Hello, how are you?'
};

chatManager.startChatStream(request, 'graph-id');

// Cancel the stream if needed
chatManager.cancelStream();

// Send feedback for a specific assistant message
// This should be called for each assistant message that receives feedback
chatManager.sendFeedback({
  user: 'user-id',
  session: 'session-id',
  query: 'Hello, how are you?',      // The user message that prompted the assistant's response
  response: 'I am doing well, thank you!', // The assistant's response being rated
  feedback: 'Great response!',        // Optional feedback text
  feedback_value: 1                   // Positive (1) or negative (-1) rating
}, 'graph-id');

React Hooks

To use the React hooks, you need to import them from the React entry point:

Callback-based Approach (Original)

import React, { useState } from 'react';
import { LoreClient, ChatRequest } from 'lore-headless';
import { useChatStream, useFeedback } from 'lore-headless/react';

// Message interface with unique ID
interface Message {
  id: string;
  text: string;
  sender: 'user' | 'assistant' | 'system' | 'error';
  timestamp: Date;
}

// Feedback modal interface
interface FeedbackModalState {
  isOpen: boolean;
  messageId: string;
  assistantMessage: string;
  userMessage: string;
}

const ChatComponent = () => {
  // Messages state
  const [messages, setMessages] = useState<Message[]>([]);
  
  // Feedback modal state
  const [feedbackModal, setFeedbackModal] = useState<FeedbackModalState>({
    isOpen: false,
    messageId: '',
    assistantMessage: '',
    userMessage: ''
  });
  
  // Feedback form state
  const [feedbackText, setFeedbackText] = useState('');
  const [feedbackRating, setFeedbackRating] = useState<number | null>(null);
  
  // Create an API client
  // Default usage (production mode)
  const apiClient = new LoreClient();
  
  // Or specify a different mode
  // const apiClient = new LoreClient('staging');
  // const apiClient = new LoreClient('development');
  
  // Or use custom URLs
  // const apiClient = new LoreClient(
  //   'production', // mode will be ignored
  //   'https://your-custom-chat-url.com',
  //   'https://your-custom-validation-url.com'
  // );
  
  // Set up authentication  
  const auth = {
    clientId: 'your-client-id',
    token: 'your-firebase-id-token'
  };
  
  // Use the chat stream hook
  const { startChatStream, cancelStream } = useChatStream(apiClient, auth, {
    onTextStream: (text) => {
      setMessages((prev) => {
        const lastMessage = prev[prev.length - 1];
        if (lastMessage?.sender === 'assistant') {
          return [
            ...prev.slice(0, -1),
            { ...lastMessage, text: lastMessage.text + text }
          ];
        }
        
        return [
          ...prev,
          {
            id: Date.now().toString(),
            text,
            sender: 'assistant',
            timestamp: new Date()
          }
        ];
      });
    },
    onError: (error) => {
      console.error('Stream error:', error);
    },
    onEnd: () => {
      console.log('Stream ended');
    }
  });
  
  // Use the feedback hook
  const { sendFeedback, loading } = useFeedback(apiClient, auth, 'graph-id');
  
  // Handle sending a message
  const handleSendMessage = (message: string) => {
    // Add user message to the list
    setMessages((prev) => [
      ...prev,
      {
        id: Date.now().toString(),
        text: message,
        sender: 'user',
        timestamp: new Date()
      }
    ]);
    
    // Start the chat stream
    const request: ChatRequest = {
      user: 'user-id',
      session: 'session-id',
      query: message
    };
    
    startChatStream(request, 'graph-id');
  };
  
  // Open feedback modal for a specific message
  const openFeedbackModal = (messageId: string) => {
    const assistantMessage = messages.find(m => m.id === messageId && m.sender === 'assistant');
    if (!assistantMessage) return;
    
    // Find the user message that prompted this assistant response
    // This is a simplified example - in a real app, you'd track which user message led to which assistant response
    const userMessageIndex = messages.findIndex(m => m.id === messageId) - 1;
    if (userMessageIndex < 0) return;
    const userMessage = messages[userMessageIndex];
    if (userMessage.sender !== 'user') return;
    
    setFeedbackModal({
      isOpen: true,
      messageId,
      assistantMessage: assistantMessage.text,
      userMessage: userMessage.text
    });
    
    // Reset form state
    setFeedbackText('');
    setFeedbackRating(null);
  };
  
  // Close feedback modal
  const closeFeedbackModal = () => {
    setFeedbackModal(prev => ({ ...prev, isOpen: false }));
  };
  
  // Submit feedback
  const submitFeedback = () => {
    if (feedbackRating === null) {
      alert('Please select a rating before submitting.');
      return;
    }
    
    sendFeedback({
      user: 'user-id',
      session: 'session-id',
      query: feedbackModal.userMessage,
      response: feedbackModal.assistantMessage,
      feedback: feedbackText || (feedbackRating > 0 ? 'Helpful response' : 'Not helpful response'),
      feedback_value: feedbackRating
    });
    
    closeFeedbackModal();
  };
  
  return (
    <div className="chat-container">
      {/* Messages list */}
      <div className="messages">
        {messages.map((message) => (
          <div key={message.id} className={`message ${message.sender}`}>
            <div className="message-content">{message.text}</div>
            <div className="message-timestamp">
              {message.timestamp.toLocaleTimeString()}
            </div>
            
            {/* Add feedback button to assistant messages */}
            {message.sender === 'assistant' && (
              <button 
                className="feedback-button"
                onClick={() => openFeedbackModal(message.id)}
              >
                Give Feedback
              </button>
            )}
          </div>
        ))}
      </div>
      
      {/* Message input form */}
      <form 
        className="input-form"
        onSubmit={(e) => {
          e.preventDefault();
          const input = e.currentTarget.elements.namedItem('message') as HTMLInputElement;
          if (input && input.value.trim()) {
            handleSendMessage(input.value.trim());
            input.value = '';
          }
        }}
      >
        <input 
          type="text" 
          name="message" 
          placeholder="Type your message..." 
          disabled={loading}
        />
        <button type="submit" disabled={loading}>Send</button>
      </form>
      
      {/* Feedback modal */}
      {feedbackModal.isOpen && (
        <div className="modal-overlay">
          <div className="modal-content">
            <div className="modal-header">
              <h3>Provide Feedback</h3>
              <button className="close-button" onClick={closeFeedbackModal}>&times;</button>
            </div>
            
            <div className="modal-body">
              <div className="message-context">
                <div className="context-label">Assistant's response:</div>
                <div className="context-text">{feedbackModal.assistantMessage}</div>
              </div>
              
              <textarea
                className="feedback-textarea"
                placeholder="Enter your feedback here..."
                value={feedbackText}
                onChange={(e) => setFeedbackText(e.target.value)}
              />
              
              <div className="rating-container">
                <div className="rating-label">How would you rate this response?</div>
                <button 
                  className={`rating-button positive ${feedbackRating === 1 ? 'selected' : ''}`}
                  onClick={() => setFeedbackRating(1)}
                >
                  👍 Helpful
                </button>
                <button 
                  className={`rating-button negative ${feedbackRating === -1 ? 'selected' : ''}`}
                  onClick={() => setFeedbackRating(-1)}
                >
                  👎 Not Helpful
                </button>
              </div>
              
              <button 
                className="submit-button"
                onClick={submitFeedback}
                disabled={loading}
              >
                Submit Feedback
              </button>
            </div>
          </div>
        </div>
      )}
    </div>
  );
};

Running the Examples

Vanilla JavaScript Example

To run the vanilla JavaScript example:

1. First, build the package:

   npm run build

2. Start the example server:

   node examples/vanilla/server.js

3. Open your browser and navigate to the URL shown in the console. The server will automatically find an available port, starting from 3000.

Note: The vanilla example uses ES modules, which require a web server to work properly due to CORS restrictions when loading modules from the file system.

React Example

The React example is provided as a reference implementation. To use it in a real React application:

1. Install the package:

   npm install path/to/lore-headless

2. Import the components and hooks as shown in the example.

Package Structure

The package is structured to be truly headless, with React-specific functionality separated into a dedicated entry point:

• Main Entry Point: Core functionality that works with any framework
• React Entry Point: React-specific hooks that depend on React

This structure allows you to use the package with or without React, depending on your needs.

Development

To develop this package:

1. Install dependencies:

   npm install

2. Build the package:

   npm run build

3. Run tests:

   npm test

API Reference

Core

LoreClient

A client for making API requests with support for different environments.

// Default usage (production mode)
const apiClient = new LoreClient();

// Specify a different mode
const stagingClient = new LoreClient('staging');
const devClient = new LoreClient('development');

// Use custom URLs (mode will be ignored)
const customClient = new LoreClient(
  'production', // This parameter will be ignored
  'https://your-custom-chat-url.com',
  'https://your-custom-validation-url.com'
);

Constructor Parameters:

• mode (string, optional): Environment mode. Options: 'production' (default), 'staging', 'development'
• customChatUrl (string, optional): Custom chat URL. Must be provided together with customValidationUrl
• customValidationUrl (string, optional): Custom validation URL. Must be provided together with customChatUrl

Environment Modes:

• Production ('production'): Uses production API endpoints
• Staging ('staging'): Uses staging API endpoints for testing
• Development ('development'): Uses local development endpoints (localhost)

Custom URLs: When custom URLs are provided, both customChatUrl and customValidationUrl must be specified together. The mode parameter will be ignored in this case.

ChatStreamManager

A manager for handling chat streams.

const chatManager = new ChatStreamManager(apiClient, auth, callbacks);

React Hooks

Message-based Approach (Enhanced)

The library now provides an enhanced version of the useChatStream hook that manages messages internally and returns them as part of the hook's return value. This simplifies message management and provides a more streamlined API.

import React from 'react';
import { LoreClient, LLMRole } from 'lore-headless';
import { useChatStream, useFeedback, ChatMessage } from 'lore-headless/react';

const ChatComponent = () => {
  const [inputText, setInputText] = useState('');
  
  // Use the enhanced useChatStream hook that returns messages
  const { 
    messages,          // Array of chat messages with different roles (user, assistant, system, error)
    isStreaming,       // Boolean indicating if a message is currently streaming
    startChatStream,   // Function to start a new chat stream
    cancelStream,      // Function to cancel the current stream
    clearMessages      // Function to clear all messages
  } = useChatStream(
    apiClient,
    auth,
    {
      // You can still use the callbacks if needed
      onSystemMessage: (text) => {
        console.log('System message:', text);
      },
    }
  );

  const handleSendMessage = async () => {
    if (!inputText.trim() || isStreaming) return;
    setInputText('');

    try {
      await startChatStream(
        {
          user: 'user-id',
          session: 'session-id',
          query: inputText,
        },
        'graph-id'
      );
    } catch (error) {
      console.error('Failed to start chat stream:', error);
    }
  };

  return (
    <div className="chat-container">
      {/* Messages list */}
      <div className="messages">
        {messages.map((message) => (
          <div 
            key={message.id} 
            className={`message ${message.role}`}
          >
            <div className="message-header">
              {message.role.charAt(0).toUpperCase() + message.role.slice(1)}
              {message.isStreaming && <span className="streaming-indicator" />}
            </div>
            <div className="message-content">{message.content}</div>
            <div className="message-timestamp">
              {message.timestamp.toLocaleTimeString()}
            </div>
          </div>
        ))}
      </div>
      
      {/* Message input form */}
      <div className="input-container">
        <input
          type="text"
          value={inputText}
          onChange={(e) => setInputText(e.target.value)}
          placeholder="Type your message..."
          disabled={isStreaming}
        />
        <button 
          onClick={handleSendMessage}
          disabled={isStreaming || !inputText.trim()}
        >
          Send
        </button>
        
        {isStreaming && (
          <button onClick={cancelStream}>
            Cancel
          </button>
        )}
        
        <button onClick={clearMessages}>
          Clear Chat
        </button>
      </div>
    </div>
  );
};

The ChatMessage interface provides a structured way to represent different types of messages:

interface ChatMessage {
  id: string;                  // Unique identifier for the message
  role: LLMRole | 'error';     // Role of the message sender (user, assistant, system, or error)
  content: string;             // Content of the message
  timestamp: Date;             // When the message was created
  isStreaming?: boolean;       // Whether the message is currently being streamed
}

Benefits of the message-based approach:

• Automatic message management (no need to manually track messages)
• Built-in support for streaming indicators
• Consistent message structure with timestamps and IDs
• Support for different message types (user, assistant, system, error)
• Simplified API with fewer lines of code

useChatStream

A hook for using the chat stream in React.

import { useChatStream } from 'lore-headless/react';

// Callback-based approach (original)
const { startChatStream, cancelStream } = useChatStream(apiClient, auth, callbacks);

// Message-based approach (enhanced)
const { messages, isStreaming, startChatStream, cancelStream, clearMessages } = useChatStream(apiClient, auth, callbacks);

useFeedback

A hook for sending feedback in React.

import { useFeedback } from 'lore-headless/react';

const { sendFeedback, loading, error, data } = useFeedback(apiClient, auth, graphId);

License

MIT