nanorag

Library Objectives

Being a lightweight option of RAG, with just the necessary dependencies.
Focused on RAG with local and open source models, not focused on API calls.
Try out different strategies and data-structures that can be used to have better results. (Such as dataframes, can try out with polars as its really performant)
Multimodal support, combine image, text and audio to get the best results
Solve some of the storage challenges RAG faces, and provide good solutions for updating documents and embeddings as well as loading them.
Use it as a educational library to demostrate on some of the main concepts llama-index or other RAG framworks use.
The base for the implementation of some agentic strategies I will try out on other library.
Offer support for local models with CPU like llama-cpp

pip install nanorag

You can take a look at the notebooks to understand how it works.