thiswillbeyourgithub/wdoc

Implements a multi-stage RAG pipeline using LangChain and LiteLLM that combines cheap and expensive LLM calls for high-recall document retrieval, then hierarchically aggregates answers via semantic batching to produce sourced markdown output with exact document citations. Supports 15+ filetypes simultaneously (PDFs, EPUBs, Anki decks, audio, video, web pages) and exposes both a CLI, Python library, and Gradio web UI for flexible integration with any LLM provider or local models.

510 stars and 840 monthly downloads. Actively maintained with 4 commits in the last 30 days. Available on PyPI.