DeepSynth

DeepSynth is a web-based chatbot powered by a local LLM (Google's DeepMind Gemma 3 4B, distilled 4-bit quantized model). It allows users to chat securely with a locally hosted language model and download their conversation logs as JSON. No external database is required — all data is handled in-memory and can be exported seamlessly from the frontend.

---

Features

• Local LLM Integration: Uses Google's DeepMind Gemma 3 4B locally via Ollama for fast, private, and secure inference.
• Modern Frontend: Built with React.js (Vite) and styled using Tailwind CSS. It features a responsive chat UI with markdown support and code highlighting.
• Robust Backend: Node.js + Express.js backend handles secure communications between the frontend and the local model.
• Authentication: Secured user login via Auth0 SPA + JWT backend, providing protected endpoints for the chat interface.
• Downloadable Chat Logs: Easily export your conversations as structured JSON files directly from the UI.
• No Database Needed: All chat sessions are temporary. You maintain control of your data by saving it locally.

---

Tech Stack

Frontend:

• React.js (Vite)
• Tailwind CSS
• React Markdown (with highlight.js support)
• Auth0 for React

Backend:

• Node.js
• Express.js
• Express OAuth2 JWT Bearer for API security
• Express Rate Limit for DDoS protection

AI & Local Execution:

• Google's DeepMind Gemma 3 4B (distilled 4-bit quantized model)
• Ollama

Deployment & Containerization:

• Docker & Docker Compose

---

Project Structure

Frontend Structure (/client)

The client is a React application built with Vite.

• /src: Contains all the React components, styling, and application logic.
  • Manages the UI for the chat interface and welcome screen.
  • Integrates Auth0 for user authentication.
  • Handles the downloading of chat sessions.
• package.json: Lists frontend dependencies like @auth0/auth0-react, tailwindcss, react-markdown, and axios.

Backend Structure (/server)

The backend is an Express API that interfaces with the local LLM. It is strictly structured into modular components:

• app.js: The main entry point initializing Express, CORS, and routing.
• /config:
  • systemPrompt.js: Defines the personality and boundaries of the AI character (DeepSynth).
• /controllers:
  • chatController.js: Manages the business logic for /chat. Sends prompts alongside the system prompt to the local Ollama instance and returns the AI's response.
• /middlewares:
  • rateLimit.js: Configures an express-rate-limit middleware to prevent abuse (DDoS protection) on the API endpoints.
• /routes:
  • chatRoute.js: Defines the POST /chat route. It applies rate limiting and JWT authentication middleware before routing the request to chatController.
• /services:
  • auth0.js: Sets up the express-oauth2-jwt-bearer middleware to validate Auth0 tokens and secure the API.

---

Screenshots

Welcome Page

Chat Interface

AI Response

Account & Login

---

Getting Started

Prerequisites

Before you begin, ensure you have the following installed on your machine:

• Node.js (v18+ recommended)
• npm or yarn
• Ollama (Ensure you have a model like gemma3:4b downloaded locally)
• Docker (optional, for containerized setup)

Installation

1. Clone the repository:

   git clone https://github.com/rishhbh/deepsynth.git
   cd deepsynth

2. Environment Setup: You will need to configure your Auth0 environment variables in both the client and server directories. Create a .env file in each directory containing the required Auth0 Domain, Client ID, and API Audience keys.

Option A: Manual Setup

Start the Frontend:

cd client
npm install
npm run dev

Start the Backend:

cd server
npm install
npm start

Option B: Docker Setup

If you prefer to run the application using Docker, you can use the provided docker-compose.yml file.

docker compose up --build

This will start both the client (on port 5173) and the server (on port 5000).

---

Future Improvements

• Light mode support for the frontend UI.
• Database integration for persistent chat history and user preferences.
• Enhanced analytics for exported conversation data.
• Support for multiple local LLM models (allowing users to select different Ollama models).
• Mobile-friendly responsive design enhancements for better mobile usability.

---

Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page if you want to contribute.

License

This project is licensed under the MIT License - see the LICENSE file for details.