tinyrag 0.3.2

A minimal Python library for Retrieval-Augmented Generation with codebase indexing and multiple vector store backends

TinyRag 🚀

A minimal, powerful Python library for Retrieval-Augmented Generation (RAG) with codebase indexing and support for multiple document formats and vector storage backends.

🌟 Features

• 🔌 Multiple Vector Stores: Faiss, ChromaDB, In-Memory, Pickle-based
• 📄 Document Support: PDF, DOCX, TXT, and raw text
• 💻 Codebase Indexing: Function-level indexing for all major programming languages
• 🧠 Default Embeddings: Uses all-MiniLM-L6-v2 by default (no API key needed)
• 🚀 Multithreading Support: Parallel document processing for faster indexing
• 🔍 Query Without LLM: Direct similarity search functionality
• 💬 Optional LLM Integration: Chat completion with retrieved context
• ⚡ Minimal Setup: Works out of the box without configuration
• 🎯 Easy to Use: Simple API with powerful features

🚀 Quick Start

Installation

# Basic installation
pip install tinyrag

# With all optional dependencies
pip install tinyrag[all]

# Specific vector stores
pip install tinyrag[faiss]    # High performance
pip install tinyrag[chroma]   # Persistent storage
pip install tinyrag[docs]     # Document processing

Usage Examples

Basic Usage (No API Key Required)

from tinyrag import TinyRag

# Initialize with default all-MiniLM-L6-v2 embeddings
rag = TinyRag()

# Add documents and codebase
rag.add_documents(["doc1.pdf", "doc2.txt", "Raw text"])
rag.add_codebase("./my_project")  # Index entire codebase

# Query documents and code
results = rag.query("What is this about?")
code_funcs = rag.search_code("authentication function")
print("Similar chunks:", results)
print("Code functions:", code_funcs)

With LLM for Chat

from tinyrag import Provider, TinyRag

provider = Provider(
    api_key="sk-xxxxxx",
    model="gpt-4",
    embedding_model="default",
    base_url="https://api.openai.com/v1"
)

rag = TinyRag(provider=provider, vector_store="faiss", max_workers=4)

rag.add_documents([
    "path/to/docs_or_raw_text",
    "Another document", 
    "More content"
])

response = rag.chat("Summarize the documents.")
print(response)

📖 Documentation

Core Components

Provider Class

Handles API interactions and embeddings:

from tinyrag import Provider

# Local embeddings only (no API key needed)
provider = Provider(embedding_model="default")

# With OpenAI API
provider = Provider(
    api_key="sk-your-key",
    model="gpt-4",
    embedding_model="text-embedding-ada-002",
    base_url="https://api.openai.com/v1"
)

TinyRag Class

Main interface for RAG operations:

from tinyrag import TinyRag

# Initialize with different vector stores
rag = TinyRag(provider, vector_store="memory")     # No dependencies
rag = TinyRag(provider, vector_store="faiss")      # High performance  
rag = TinyRag(provider, vector_store="chroma")     # Persistent
rag = TinyRag(provider, vector_store="pickle")     # Simple file-based

Vector Store Comparison

Store	Performance	Persistence	Memory	Dependencies	Best For
Memory	Good	Manual	High	None	Development, small datasets
Faiss	Excellent	Manual	Low	faiss-cpu	Large-scale, performance-critical
ChromaDB	Good	Automatic	Medium	chromadb	Production, automatic persistence
Pickle	Fair	Manual	Medium	scikit-learn	Simple file-based storage

API Reference

Core Methods

# Document Management
rag.add_documents(data)                    # Add documents/text
rag.get_chunk_count()                      # Get number of chunks
rag.get_all_chunks()                       # Get all text chunks
rag.clear_documents()                      # Clear all data

# Codebase Indexing
rag.add_codebase(path)                     # Index codebase at function level
rag.search_code(query, k=5, language=None) # Search code functions
rag.get_function_by_name(name, k=3)        # Find functions by name

# Querying (No LLM)
rag.query(query, k=5, return_scores=True) # Basic similarity search
rag.search_documents(query, k=5, min_score=0.0) # With score filtering
rag.get_similar_chunks(text, k=5)         # Find similar to given text

# LLM Integration
rag.chat(query, k=3)                      # Generate response with context

# Persistence
rag.save_vector_store(filepath)           # Save to disk
rag.load_vector_store(filepath)           # Load from disk

Codebase Indexing

TinyRag can automatically parse and index codebases at the function level:

Supported Languages

• Python (.py)
• JavaScript/TypeScript (.js, .ts)
• Java (.java)
• C/C++ (.c, .cpp, .cc, .cxx, .h)
• Go (.go)
• Rust (.rs)
• PHP (.php)

Usage Examples

from tinyrag import TinyRag

rag = TinyRag()

# Index entire codebase
rag.add_codebase("./my_project", recursive=True)

# Search for specific functionality
auth_functions = rag.search_code("authentication login", k=5)

# Search by programming language
python_funcs = rag.search_code("database query", language="python")

# Find specific function
user_func = rag.get_function_by_name("create_user")

# Code-aware chat (with API key)
response = rag.chat("How does the authentication system work?")

🔧 Configuration Options

Vector Store Configuration

# Faiss with custom settings
rag = TinyRag(
    provider=provider,
    vector_store="faiss",
    chunk_size=1000,  # Larger chunks
    vector_store_config={}
)

# ChromaDB with persistence
rag = TinyRag(
    provider=provider,
    vector_store="chroma", 
    vector_store_config={
        "collection_name": "my_collection",
        "persist_directory": "./chroma_db"
    }
)

# Memory store (no config needed)
rag = TinyRag(provider=provider, vector_store="memory")

# Pickle store with scikit-learn
rag = TinyRag(provider=provider, vector_store="pickle")

Provider Configuration

# Local embeddings only
provider = Provider(embedding_model="default")

# OpenAI with custom settings
provider = Provider(
    api_key="sk-your-key",
    model="gpt-3.5-turbo",
    embedding_model="text-embedding-ada-002",
    base_url="https://api.openai.com/v1"
)

# Custom API endpoint
provider = Provider(
    api_key="your-key",
    model="custom-model",
    base_url="https://your-custom-api.com/v1"
)

📦 Installation Options

# Minimal installation
pip install tinyrag

# With specific vector stores
pip install tinyrag[faiss]      # For high-performance similarity search
pip install tinyrag[chroma]     # For persistent vector database
pip install tinyrag[pickle]     # For simple file-based storage

# With document processing
pip install tinyrag[docs]       # PDF and DOCX support

# Everything included
pip install tinyrag[all]        # All optional dependencies

🛠️ Development

Requirements

• Python 3.7+
• sentence-transformers (core)
• requests (core)
• numpy (core)

Optional Dependencies

• faiss-cpu: High-performance vector search
• chromadb: Persistent vector database
• scikit-learn: Pickle vector store similarity
• PyPDF2: PDF document processing
• python-docx: Word document processing

Contributing

1. Fork the repository: https://github.com/Kenosis01/TinyRag.git
2. Create a feature branch: git checkout -b feature-name
3. Make your changes and add tests
4. Submit a pull request

📄 License

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

🤝 Support

• GitHub Issues: Report bugs or request features
• Documentation: Full documentation
• Examples: Check the examples/ directory in the repository

🎯 Use Cases

• Document Q&A: Query your documents without LLM costs
• Knowledge Base: Build searchable knowledge repositories
• Content Discovery: Find similar content in large document collections
• RAG Applications: Full retrieval-augmented generation workflows
• Research Tools: Semantic search through research papers
• Customer Support: Query company documentation and policies

---

TinyRag - Making RAG simple, powerful, and accessible! 🚀