RAGSearch

A minimal Retrieval-Augmented Generation search tool. Point it at a folder of .txt / .md / .pdf files and ask questions; answers are generated by an OpenAI model using only your documents as context, with inline citations.

Setup

# 1. Install dependencies (a virtualenv is recommended)
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# 2. Add your OpenAI API key
cp .env.example .env
# then edit .env and paste your key after OPENAI_API_KEY=

Get a key at https://platform.openai.com/api-keys

Usage

# Drop your PDFs (or .txt/.md files) into the ./data folder, then build the index.
# You can pass a folder or a single file:
python main.py ingest data
python main.py ingest data/my-report.pdf

# Ask a one-off question
python main.py ask "How does RAGSearch work?"

# Or chat interactively
python main.py chat

How it works

Step	What happens
ingest	Files are chunked, each chunk is embedded with text-embedding-3-small, and vectors are saved to a local index.npz / index.json.
ask	Your question is embedded, compared to every chunk via cosine similarity, and the top matches are sent to gpt-4o-mini as grounding context.

Configuration

Set these in .env to override the defaults:

• OPENAI_API_KEY — required
• EMBEDDING_MODEL — default text-embedding-3-small
• CHAT_MODEL — default gpt-4o-mini

Your .env and the generated index are git-ignored.