⚡ PocketRAG

Privacy-first, local RAG engine. Run LLM search on your documents without the cloud.

PocketRAG turns your local documents into an intelligent, searchable database. Built for speed, privacy, and minimal resource usage (optimized for Qwen3.5 0.8B).

🚀 Why PocketRAG?

• 100% Offline: Your data never leaves your machine.
• Ultra-Light: Designed to run on consumer laptops (no massive GPUs required).
• Zero-Setup: Single-command CLI.
• Smart Retrieval: Uses vector embeddings to find exact answers, not just keywords.
• Production Ready: Modular architecture, proper error handling, and logging.

🛠️ Quick Start

Installation

# Clone and install
git clone https://github.com/Qeuph/PocketRag
cd pocketrag
pip install -e .

# Or install directly from PyPI (when published)
# pip install pocketrag

Usage

# 1. Initialize the database
pocketrag init

# 2. Add documents to index
pocketrag add ./my_documents

# 3. Search documents
pocketrag search "your query here"

# 4. Chat with your documents
pocketrag chat

# Check status anytime
pocketrag status

Advanced Usage

# Index non-recursively
pocketrag add ./docs --no-recursive

# Use a different model for chat
pocketrag chat --model llama3.2:1b

# Adjust retrieval count
pocketrag chat --top-k 10

# Search with custom result count
pocketrag search "query" --top-k 3

# Clear all indexed data
pocketrag clear --yes

🏗️ Architecture

PocketRAG uses a modular, production-ready architecture:

pocketrag/
├── config/          # Configuration management
├── core/            # Core RAG components
│   ├── chunker.py   # Text chunking strategies
│   ├── embedding.py # Embedding generation
│   ├── indexer.py   # Document indexing pipeline
│   ├── search.py    # Vector search & retrieval
│   └── chat.py      # LLM chat integration
├── utils/           # Utility functions
│   └── document_loader.py  # Multi-format document parsing
└── cli.py           # Command-line interface

Components

Component	Technology	Purpose
Vector Store	LanceDB	Serverless vector storage
Embeddings	SentenceTransformers	High-performance local embeddings
LLM	Ollama	Local LLM inference
CLI	Typer + Rich	Beautiful command-line interface

📁 Supported File Formats

• Documents: PDF, TXT, MD
• Code: Python (.py), JavaScript (.js), TypeScript (.ts/.tsx), Java (.java), C/C++ (.c/.cpp/.h), Go (.go), Rust (.rs), Ruby (.rb), PHP (.php)
• Web: HTML, CSS, JSON

🔧 Configuration

PocketRAG uses sensible defaults but can be customized:

from pocketrag import Config, Indexer, ChatEngine

# Custom configuration
config = Config(
    db_path="./custom_db",
    chunk_size=1000,
    chunk_overlap=100,
    embedding_model="all-MiniLM-L6-v2",
    default_top_k=5,
)

# Programmatic usage
indexer = Indexer(
    db_path=config.db_path,
    chunk_size=config.chunk_size,
)
indexer.index_directory("./docs")

chat = ChatEngine(model_name="llama3.2:1b")
response = chat.chat("What's in my documents?")

🤝 Contributing

Built for developers who value privacy. PRs are welcome!

Development Setup

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black pocketrag/

# Lint
ruff check pocketrag/

📄 License

MIT License - See LICENSE for details.

---

If this project saved you time, please ⭐ star the repo!