Skip to main content

​
Welcome to Mini RAG 🚀

Mini RAG is a lightweight, modular, and production-ready Retrieval-Augmented Generation (RAG) library built with Python. Install with uv add mini-rag and start building intelligent document search and question-answering systems in minutes.

​
Key Features

🤖 Agentic RAG

Intelligent query processing with automatic query rewriting and result re-ranking

📄 Multi-format Support

Load documents from PDF, DOCX, images, and more using MarkItDown

✂️ Smart Chunking

Advanced text chunking with Chonkie for optimal context preservation

🔮 Flexible Embeddings

Support for OpenAI, Azure OpenAI, and any OpenAI-compatible API

💾 Vector Storage

Powered by Milvus for high-performance similarity search

🎯 Query Optimization

Automatic query rewriting for better retrieval results

🔍 Hybrid Search

Combine semantic (vector) and keyword (BM25) search

📊 Multiple Re-ranking

Choose from Cohere API, local cross-encoders, or LLM-based re-ranking

📈 Observability

Built-in Langfuse integration for tracing and monitoring

🔧 Modular Design

Use individual components or the complete RAG pipeline

​
Quick Start

Get started with Mini RAG in just 5 lines of code: 
import os
from mini import AgenticRAG, EmbeddingModel, VectorStore

# Initialize
embedding_model = EmbeddingModel()
vector_store = VectorStore(
    uri=os.getenv("MILVUS_URI"),
    token=os.getenv("MILVUS_TOKEN"),
    collection_name="my_docs",
    dimension=1536
)
rag = AgenticRAG(vector_store=vector_store, embedding_model=embedding_model)

# Index a document
rag.index_document("path/to/document.pdf")

# Ask a question
response = rag.query("What is this document about?")
print(response.answer)

Installation

Install Mini RAG and get set up in minutes

Quick Start Guide

Follow our step-by-step guide to build your first RAG application

API Reference

Explore the complete API documentation

Examples

Learn from practical examples and use cases

​
Why Mini RAG?

Mini RAG provides a clean, intuitive API that follows Python best practices. Get started with just a few lines of code.
Built with production use cases in mind, featuring error handling, retry logic, observability, and comprehensive configuration options.
Use individual components (loader, chunker, embeddings, vector store) or the complete RAG pipeline. Mix and match as needed.
Query rewriting, hybrid search, multiple re-ranking strategies, and observability built-in—features that typically require custom implementation.
Support for multiple embedding providers, vector stores, and re-ranking methods. Easy to extend with custom implementations.

​
Architecture

Mini RAG follows a modular architecture that makes it easy to understand and customize: 
┌─────────────────────────────────────────────────────────────┐
│                      AgenticRAG System                       │
└─────────────────────────────────────────────────────────────┘
                              │
        ┌─────────────────────┼─────────────────────┐
        │                     │                     │
        ▼                     ▼                     ▼
┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│ DocumentLoader│    │   Chunker    │    │EmbeddingModel│
│  (MarkItDown) │───▶│  (Chonkie)   │───▶│   (OpenAI)   │
└──────────────┘    └──────────────┘    └──────────────┘
                                                 │
                                                 ▼
                                        ┌──────────────┐
                                        │ VectorStore  │
                                        │   (Milvus)   │
                                        └──────────────┘

​
Community & Support

GitHub

Star us on GitHub and contribute

PyPI

View releases and version history

Issues

Report bugs or request features

​
Next Steps

1

Install Mini RAG

Follow the installation guide to set up Mini RAG
2

Complete Quick Start

Build your first RAG application with our quick start guide
3

Explore Features

Learn about advanced features like hybrid search and re-ranking
4

Check Examples

Browse practical examples for your use case