ChatGPT Clone

A ChatGPT-like web interface powered by open-source LLMs running locally via KDeps.

Features

• Modern, dark-themed UI similar to ChatGPT
• Multiple model selection (Llama, Mistral, Phi, Gemma, Qwen)
• Persistent chat history (saved in browser localStorage)
• Real-time loading indicators
• Markdown-like formatting (code blocks, bold, italic)
• Fully local - your data never leaves your machine

Quick Start

Prerequisites

1. Install KDeps
2. Install Ollama and ensure it's running
3. Pull at least one model:

   ollama pull llama3.2:1b
   

Run Locally

# Run from the project root
kdeps run examples/chatgpt-clone

This starts both servers:

• API Server on http://localhost:16395 (handles chat requests)
• Web Interface on http://localhost:16395 (serves the frontend)

Open http://localhost:16395 in your browser to use the chat interface.

Architecture

examples/chatgpt-clone/
├── workflow.yaml           # Main workflow configuration
├── resources/
│   ├── llm.yaml           # LLM chat handler
│   ├── models.yaml        # Available models endpoint
│   └── response.yaml      # Response formatting
└── public/
    ├── index.html         # Web interface
    ├── styles.css         # Dark theme styles
    └── app.js             # Frontend JavaScript

API Endpoints

POST /api/v1/chat

Send a chat message to the LLM.

Request:

{
  "message": "What is AI?",
  "model": "llama3.2:1b"
}

Response:

{
  "success": true,
  "response": {
    "message": "AI (Artificial Intelligence) is...",
    "model": "llama3.2:1b",
    "query": "What is AI?"
  }
}

GET /api/v1/models

Get list of available models.

Response:

{
  "success": true,
  "response": {
    "models": [
      {"id": "llama3.2:1b", "name": "Llama 3.2 (1B)", "description": "..."},
      ...
    ]
  }
}

Available Models

Model	Size	Description
llama3.2:1b	1B	Meta's smallest and fastest model
llama3.2:3b	3B	Balanced speed and quality
mistral:7b	7B	Excellent performance
phi3:3.8b	3.8B	Microsoft's efficient model
gemma2:2b	2B	Google's lightweight model
qwen2.5:3b	3B	Multilingual support

To add more models, pull them with Ollama:

ollama pull codellama:7b
ollama pull deepseek-coder:6.7b

Then add them to the model selector in public/index.html.

Docker Deployment

Build a Docker image with pre-loaded models:

kdeps build workflow.yaml

The resulting image will include:

• Ollama with pre-loaded models
• KDeps runtime
• Static web interface

Run the container:

docker run -p 16395:16395 -p 16395:16395 chatgpt-clone:latest

Customization

Change Default Model

Edit resources/llm.yaml:

expr:
  - set('selectedModel', default(get('model'), 'mistral:7b'))

Add System Prompt

Edit resources/llm.yaml:

expr:
  - set('systemPrompt', 'You are a coding assistant. Always provide examples.')

Change UI Theme

Edit public/styles.css to customize colors, fonts, and layout.

Troubleshooting

"Unable to connect to the server"

1. Check if KDeps is running: kdeps run workflow.yaml
2. Check if Ollama is running: ollama list
3. Verify the API: curl http://localhost:16395/api/v1/models

"Model not found"

Pull the model with Ollama:

ollama pull llama3.2:1b

Slow responses

• Use a smaller model (1B or 3B)
• Ensure you have enough RAM (8GB+ recommended)
• For GPU acceleration, install Ollama with CUDA support

License

MIT License - See LICENSE for details.