@gilf/chat-websocket-server NPM

WebSocket Chat Server with LLM Integration

A TypeScript-based WebSocket server that enables chat functionality with both AI language models (ChatGPT and Gemini) and human operators. The system allows users to chat with AI models when no operator is available, and seamlessly switch to human operator support when needed.

Features

Real-time chat using WebSocket connections
Support for multiple AI providers:
- OpenAI's ChatGPT
- Google's Gemini
Human operator integration:
- Secure operator authentication
- One-to-one chat sessions between operators and users
- Operator can only chat with one user at a time
Automatic fallback to AI when no operator is available
Command-line interfaces for both users and operators
Written in TypeScript for better type safety and developer experience

Prerequisites

Node.js (v16 or higher)
Yarn package manager
OpenAI API key
Google Gemini API key

Installation

Clone the repository:

git clone https://github.com/onepointconsulting/operator-chat
cd chat-websocket-server

Install dependencies:

yarn install

Create a .env file in the root directory with the following variables:

OPENAI_API_KEY=<key>
OPENAI_MODEL=gpt-4o-mini
INITIAL_PROVIDER=openai
# INITIAL_PROVIDER=gemini # uncomment this to use gemini
GEMINI_API_KEY=<key>
GEMINI_MODEL=gemini-1.5-flash

OPERATOR_PASSWORD=<pass>
PORT=4000

PROMPT_FILE=config/prompts.toml
SLICE_SIZE=5

You can use the .env_local file as a template to create your own .env file.

Make sure you have the prompts file in the config folder.

Here is an example of the prompts file:

[basic]
system_message = """You are a helpful assistant in british English.

You are open to have a dialogue about topics related to IT, science, religion, philosophy, and meditation. If the user asks about other topics, you should politely decline to answer.

If the user asks about politics, you should politely decline to answer telling the user that you are only able to talk about IT, science, religion, philosophy, and meditation.

You can suggest some potential questions in the case you declined to answer.

"""

initial_questions = [
    "What is your name?",
    "What is your role in the company?"
]

[configuration]
max_history_size = 5