Repolode
Submit
thiswillbeyourgithub

thiswillbeyourgithub/wdoc

A RAG tool that summarizes and answers questions across mixed documents (PDFs, audio, YouTube, Anki, Logseq, web pages) using any LLM provider.

View on GitHubWebsite
Stars
549
Forks
47
Open issues
2
Last commit
10 days ago

What is it?

wdoc is a Python RAG system for summarizing and querying a large, mixed collection of documents at once. It can load PDFs, EPUBs, Office files, audio and video (transcribed with Whisper or Deepgram), YouTube videos and playlists, web pages, Anki collections, Logseq graphs, and more, and combine them all in one index. Answers are returned as markdown with references to the source document for each claim.

The query pipeline retrieves candidate chunks by embedding search, uses a cheaper LLM to filter out irrelevant documents, answers from each remaining document with a stronger LLM, and then merges those answers in semantically clustered batches. Summary mode processes text chunk by chunk and produces indented bullet points that aim to capture the author's reasoning. LLM access goes through LiteLLM, so OpenAI, Anthropic, Mistral, Ollama and other providers can be used, and a private mode disables external API keys and outgoing connections.

wdoc can be run from the command line, used as a Python library, or deployed through an experimental Docker Gradio web UI. Optional extras keep the base install small, and the project also includes an Open-WebUI tool.

Key features

  • 01Loads 15+ file types, including PDFs, EPUBs, Office files, audio, video, YouTube, Anki and Logseq.
  • 02Combines multiple sources into one index and queries them together.
  • 03Returns sourced markdown answers that point to the originating document.
  • 04Works with any LLM provider supported by LiteLLM, including local models via Ollama.
  • 05Offers a private mode that blocks outgoing connections and API keys.
  • 06Provides summarization that preserves the author's arguments as indented bullets.

Use cases

  • Asking questions across a personal library of PDFs, notes and saved articles.
  • Summarizing long YouTube videos, podcasts or lectures into structured notes.
  • Querying an Anki flashcard collection alongside other study materials.
  • Running document Q&A on sensitive files with a local LLM in private mode.

Quick start

  1. 1

    Run wdoc with uvx to search the web and get a sourced answer

    uvx wdoc web "How is Nvidia today this month?"
  2. 2

    Query a PDF with the full extras bundle (replace the link with your own)

    uvx wdoc[full] --path=$link --task=query --filetype="online_pdf" --query="What does it say about alphago?" --query_retrievers='basic_multiquery' --top_k=auto_200_500

Things to know

  • Licensed under AGPL-3.0, a copyleft license that requires derivative network services to share their source code.
  • Still under active development and described by its author as usable but with many planned features and known bugs.
  • Default models are hosted via OpenRouter, so typical use needs a paid API key unless you configure a local LLM.
  • The plain install only covers PDF and web loaders; other filetypes need the [full] extras.

In collections

PDF, OCR & document tools

Topics

#rag#llm#document-qa#summarization#pdf#langchain#litellm#python#news#parser

FAQ

What is wdoc?
A RAG tool that summarizes and answers questions across mixed documents (PDFs, audio, YouTube, Anki, Logseq, web pages) using any LLM provider. wdoc is a Python RAG system for summarizing and querying a large, mixed collection of documents at once. It can load PDFs, EPUBs, Office files, audio and video (transcribed with Whisper or Deepgram), YouTube videos and playlists, web pages, Anki collections, Logseq graphs, and more, and combine them all in one index. Answers are returned as markdown with references to the source document for each claim.
Is wdoc free to use? What license does it have?
wdoc is open source under the AGPL-3.0 license. Check the license terms before using it in commercial products.
Is wdoc still maintained?
Actively maintained. The last commit to wdoc was on Oct 28, 2026.
What can I use wdoc for?
Asking questions across a personal library of PDFs, notes and saved articles. Summarizing long YouTube videos, podcasts or lectures into structured notes. Querying an Anki flashcard collection alongside other study materials. Running document Q&A on sensitive files with a local LLM in private mode.
How do I install wdoc?
Run wdoc with uvx to search the web and get a sourced answer: uvx wdoc web "How is Nvidia today this month?" Query a PDF with the full extras bundle (replace the link with your own): uvx wdoc[full] --path=$link --task=query --filetype="online_pdf" --query="What does it say about alphago?" --query_retrievers='basic_multiquery' --top_k=auto_200_500
Where can I get wdoc?
The source code and installation instructions are on GitHub: https://github.com/thiswillbeyourgithub/wdoc

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects