What it does

KnowledgeGPT is a local web app for document Q&A. Upload a file, ask questions, and it returns answers with citations pulled from the text. The stack is deliberately simple: Streamlit for the UI, LangChain for the retrieval chain, and OpenAI’s API for the language model. Docker support is included if you’d rather not wrestle with Python environments.

The interesting bit

This was built early in the RAG boom—before “RAG” became conference-slide filler—and the README still carries that straightforward energy. The roadmap is refreshingly honest about what’s missing: OCR, more file formats, local LLMs, and actual highlighted citations rather than just text snippets. It’s a snapshot of what document AI looked like when the pieces were still being glued together in public.

Key highlights

• Citation-backed answers, not just LLM hallucinations with confidence
• 25 MB upload limit in the hosted version, configurable locally
• Docker one-liner for deployment
• Poetry-managed Python ≥3.10 environment
• MIT licensed, with contributor tracking via contrib.rocks

Caveats

• Requires a paid OpenAI API key; the README warns about rate limits on free tiers
• No local LLM support yet—it’s on the roadmap, not in the code
• The demo link is a Twitter video, which may or may not still load

Verdict

Good for developers who want a clean, hackable RAG starter they can run locally and extend. Skip it if you need production multi-tenancy or are allergic to Streamlit’s single-threaded simplicity.