ragpackai 📦

Portable Retrieval-Augmented Generation Library

ragpackai is a Python library for creating, saving, loading, and querying portable RAG (Retrieval-Augmented Generation) packs. It allows you to bundle documents, embeddings, vectorstores, and configuration into a single .rag file that can be easily shared and deployed across different environments.

✨ Features

• 🚀 Portable RAG Packs: Bundle everything into a single .rag file
• 🔄 Provider Flexibility: Support for OpenAI, Google, Groq, Cerebras, and HuggingFace
• 🔒 Encryption Support: Optional AES-GCM encryption for sensitive data
• 🎯 Runtime Overrides: Change embedding/LLM providers without rebuilding
• 📚 Multiple Formats: Support for PDF, TXT, MD, and more
• 🛠️ CLI Tools: Command-line interface for easy pack management
• 🔧 Lazy Loading: Efficient dependency management with lazy imports

🚀 Quick Start

Installation

# Core installation
pip install ragpackai

# With optional providers
pip install ragpackai[google]     # Google Vertex AI
pip install ragpackai[groq]       # Groq
pip install ragpackai[cerebras]   # Cerebras
pip install ragpackai[all]        # All providers

Basic Usage

from ragpackai import ragpackai

# Create a pack from documents
pack = ragpackai.from_files([
    "docs/manual.pdf", 
    "notes.txt",
    "knowledge_base/"
])

# Save the pack
pack.save("my_knowledge.rag")

# Load and query
pack = ragpackai.load("my_knowledge.rag")

# Simple retrieval (no LLM)
results = pack.query("How do I install this?", top_k=3)
print(results)

# Question answering with LLM
answer = pack.ask("What are the main features?")
print(answer)

Provider Overrides

# Load with different providers
pack = ragpackai.load(
    "my_knowledge.rag",
    embedding_config={
        "provider": "google", 
        "model_name": "textembedding-gecko"
    },
    llm_config={
        "provider": "groq", 
        "model_name": "mixtral-8x7b-32768"
    }
)

answer = pack.ask("Explain the architecture")

🛠️ Command Line Interface

Create a RAG Pack

# From files and directories
ragpackai create docs/ notes.txt --output knowledge.rag

# With custom settings
ragpackai create docs/ \
  --embedding-provider openai \
  --embedding-model text-embedding-3-large \
  --chunk-size 1024 \
  --encrypt-key mypassword

Query and Ask

# Simple retrieval
ragpackai query knowledge.rag "How to install?"

# Question answering
ragpackai ask knowledge.rag "What are the requirements?" \
  --llm-provider openai \
  --llm-model gpt-4o

# With provider overrides
ragpackai ask knowledge.rag "Explain the API" \
  --embedding-provider google \
  --embedding-model textembedding-gecko \
  --llm-provider groq \
  --llm-model mixtral-8x7b-32768

Pack Information

ragpackai info knowledge.rag

🏗️ Architecture

.rag File Structure

A .rag file is a structured zip archive:

mypack.rag
├── metadata.json          # Pack metadata
├── config.json           # Default configurations
├── documents/            # Original documents
│   ├── doc1.txt
│   └── doc2.pdf
└── vectorstore/          # Chroma vectorstore
    ├── chroma.sqlite3
    └── ...

Supported Providers

Embedding Providers:

• openai: text-embedding-3-small, text-embedding-3-large
• huggingface: all-MiniLM-L6-v2, all-mpnet-base-v2 (offline)
• google: textembedding-gecko

LLM Providers:

• openai: gpt-4o, gpt-4o-mini, gpt-3.5-turbo
• google: gemini-pro, gemini-1.5-flash
• groq: mixtral-8x7b-32768, llama2-70b-4096
• cerebras: llama3.1-8b, llama3.1-70b

📖 API Reference

ragpackai Class

ragpackai.from_files(files, embed_model="openai:text-embedding-3-small", **kwargs)

Create a RAG pack from files.

Parameters:

• files: List of file paths or directories
• embed_model: Embedding model in format "provider:model"
• chunk_size: Text chunk size (default: 512)
• chunk_overlap: Chunk overlap (default: 50)
• name: Pack name

ragpackai.load(path, embedding_config=None, llm_config=None, **kwargs)

Load a RAG pack from file.

Parameters:

• path: Path to .rag file
• embedding_config: Override embedding configuration
• llm_config: Override LLM configuration
• reindex_on_mismatch: Rebuild vectorstore if dimensions mismatch
• decrypt_key: Decryption password

pack.save(path, encrypt_key=None)

Save pack to .rag file.

pack.query(question, top_k=3)

Retrieve relevant chunks (no LLM).

pack.ask(question, top_k=4, temperature=0.0)

Ask question with LLM.

Provider Wrappers

# Direct provider access
from ragpackai.embeddings import OpenAI, HuggingFace, Google
from ragpackai.llms import OpenAIChat, GoogleChat, GroqChat

# Create embedding provider
embeddings = OpenAI(model_name="text-embedding-3-large")
vectors = embeddings.embed_documents(["Hello world"])

# Create LLM provider
llm = OpenAIChat(model_name="gpt-4o", temperature=0.7)
response = llm.invoke("What is AI?")

🔧 Configuration

Environment Variables

# API Keys
export OPENAI_API_KEY="your-key"
export GOOGLE_CLOUD_PROJECT="your-project"
export GROQ_API_KEY="your-key"
export CEREBRAS_API_KEY="your-key"

# Optional
export GOOGLE_APPLICATION_CREDENTIALS="path/to/service-account.json"

Configuration Files

# Custom embedding config
embedding_config = {
    "provider": "huggingface",
    "model_name": "all-mpnet-base-v2",
    "device": "cuda"  # Use GPU
}

# Custom LLM config
llm_config = {
    "provider": "openai",
    "model_name": "gpt-4o",
    "temperature": 0.7,
    "max_tokens": 2000
}

🔒 Security

Encryption

ragpackai supports AES-GCM encryption for sensitive data:

# Save with encryption
pack.save("sensitive.rag", encrypt_key="strong-password")

# Load encrypted pack
pack = ragpackai.load("sensitive.rag", decrypt_key="strong-password")

Best Practices

• Use strong passwords for encryption
• Store API keys securely in environment variables
• Validate .rag files before loading in production
• Consider network security when sharing packs

🧪 Examples

See the examples/ directory for complete examples:

• basic_usage.py - Simple pack creation and querying
• provider_overrides.py - Using different providers
• encryption_example.py - Working with encrypted packs
• cli_examples.sh - Command-line usage examples

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

📄 License

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

🆘 Support

• 📖 Documentation
• 🐛 Issue Tracker
• 💬 Discussions

🙏 Acknowledgments

Built with:

• LangChain - LLM framework
• ChromaDB - Vector database
• Sentence Transformers - Embedding models