Notes Chat

This is a small conversational RAG app for asking questions over text documents. I use it for my markdown notes, but it should work with any text-based documents. PDF support is in the works.

The current flow is:

1. Sync markdown notes from ditek/Notes or use local files in notes/.
2. Split notes into chunks with LangChain's text splitter.
3. Embed chunks with Hugging Face.
4. Store vectors in Chroma.
5. Retrieve relevant chunks for each question.
6. Rewrite follow-up questions using chat history.
7. Answer with a Hugging Face chat model.

Setup

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Edit .env and set a Hugging Face token with at least this permission:

Make calls to Inference Providers

Sync Notes

Download notes a public git repo (specified in NOTES_REPO_CONTENTS_URL):

python main.py sync-notes

Download and rebuild the vector index:

python main.py sync-notes --index

You can also index whatever markdown files are already in notes/:

python main.py index --reset

Ask From The CLI

python main.py ask "What is an Ansible inventory?" --k 3

Run The Streamlit App

streamlit run app.py

The app includes:

• chat history
• query rewriting for follow-up questions
• source chunk display
• a k slider for retrieved chunks
• buttons to sync notes, rebuild the index, reload the index, and clear chat

Environment Variables

• Set HF_TOKEN to allow communication with Hugging Face. Only Make calls to Inference Providers permission is needed for that.
• Set CHROMA_DIR=./chroma_db for local development.
• Keep ENABLE_SIDEBAR_CONTROLS=false to hide the sidebar in public embeds.
• Set DEFAULT_RETRIEVAL_K to control how many chunks are retrieved when the sidebar is hidden.

Deployments

Deployment-specific files live under deploy/.

For deploying to Hugging Face Spaces, see:

deploy/huggingface/README.md