// the find
hwchase17/notion-qa
A 2022-era demo by LangChain's creator showing how to do Q&A over a Notion export using embeddings and a retrieval chain. It converts your exported Markdown files into a local vector store, then lets you query it in plain English. Useful as a learning example; not a tool you'd run in production.
Straightforward two-script setup — ingest.py builds the index, qa.py queries it, no magic. The offline-export approach sidesteps Notion API auth headaches entirely. Comes with real sample data (Blendle handbook) so you can run it without touching your own workspace first. Good minimal example of the embed-then-retrieve pattern that LangChain is built around.
Completely unmaintained — last touched September 2024, and the LangChain APIs it uses have broken repeatedly across major versions, so it almost certainly doesn't run against current langchain packages without fixing imports. No incremental sync: every update to your Notion workspace means re-exporting and re-ingesting the whole thing. The export zip is checked into the repo itself, which is a bad habit for anything with sensitive content. No conversation memory — each question is stateless, so follow-up questions don't work.