pirag 0.2.2

CLI Projects of On-Premise RAG. You can use your own LLM and vector DB. Or just add remote LLM servers and vector DB.

pirag: pilot-onpremise-rag

🌱 LLM+RAG CLI project operating in On-Premise environment

🚀 Introduction

pilot-onpremise-rag is a CLI tool that implements a knowledge-based RAG (Retrieval-Augmented Generation) system with LLM. It provides powerful document retrieval and generation capabilities while ensuring data privacy.

🔧 Setup

Install pirag from PyPI

pip install pirag

Install pirag from source

git clone https://github.com/jyje/pilot-onpremise-rag
cd pilot-onpremise-rag

pip install --upgrade -e .

(Optional) Setup External Dependencies

git clone https://github.com/jyje/pilot-onpremise-rag
cd pilot-onpremise-rag

docker compose -f docker/compose.yaml up

📚 Usage

Basic Commands

# View help
pirag --help

# Train documents
pirag train --source ./documents

# Ask a question
pirag ask "Give me a joke for Cat-holic."

🏗️ Project Structure

pilot-onpremise-rag/
├── app/                        # Main application directory
│   ├── main.py                 # CLI main entry point
│   ├── setup.py                # Package setup configuration
│   ├── pyproject.toml          # PEP 517/518 build configuration
│   ├── requirements.txt        # Dependencies
│   ├── logs/                   # Application logs
│   └── rag/                    # RAG implementation
│       ├── config.py           # Configuration settings
│       ├── agent.py            # Agent implementation
│       ├── ask/                # Query handling module
│       ├── train/              # Document training module
│       ├── test/               # Testing module
│       └── doctor/             # Diagnostic tools
├── VERSION                     # Project version
├── docker/                     # Docker configuration
├── assets/                     # Static assets (Files are not included)
└── LICENSE                     # License information

🔄 How It Works

1. Document Training: Process local documents and store in vector database
2. Search Engine: Find document chunks related to user queries
3. Context Generation: Create LLM prompts from retrieved documents
4. Response Generation: Provide accurate responses via local LLM

💡 Key Features

• Privacy Guaranteed: All data and processing occurs locally
• Multiple Document Support: Support for PDF, Markdown, TXT, DOCX, and other formats
• Custom LLM: Compatible with various local LLM models
• Vector Database: Vector DB integration for efficient document retrieval

🧪 Performance Optimization

Configuration	Memory Usage	Response Speed	Suitable Use Cases
Light Model	4-6GB	Fast	Simple queries, low-spec hardware
Medium Model	8-12GB	Medium	General use, most queries
Large Model	16GB+	Slow	Complex document analysis, expert answers

🔗 References

• LangChain Documentation
• LLM Optimization Techniques
• RAG Paper

Contributing

Any contributions are welcome!

Current Maintainers

• Studio R4iny
  • jyje, semir4in (Same person)