Most "chat with your docs" demos fall apart the moment you ask where an answer came from: they paraphrase a PDF and leave you to trust them. The bet here is that a document-QA tool is only useful if every claim is traceable, so each answer is paired with an in-browser PDF viewer that highlights the exact passage it was pulled from. It also ships as two things at once: a finished multi-user app and a pipeline framework you can rebuild.
What Sets It Apart
- Citation-grounded answers: the viewer highlights the source passage, so you verify each claim instead of trusting a paraphrase.
- Two products in one repo: a clean multi-user UI (login, private/public collections) for non-technical users, plus a Gradio-based RAG framework developers can extend.
- Hybrid retrieval and multi-modal parsing: full-text plus vector search, with OCR, table, and figure extraction, so scanned or table-heavy PDFs don't degrade to noise.
- Provider-agnostic with advanced reasoning: OpenAI, Azure, Cohere, Ollama, or fully local models, plus question decomposition, agent modes (ReAct, ReWOO), and optional GraphRAG.
Great Fit If
You want a self-hostable, Apache-2.0 document-QA app you can run locally and later customize, or a clean starting point for building your own RAG pipeline. Look elsewhere if you need a turnkey managed SaaS with zero operations — this is a Python 3.10+/Docker app you host yourself — or if your need is simple single-file Q&A where a hosted assistant is far less setup.
