RAQA

Retrieval-Augmented Question-Answering

Retrieval-augmented, pip-installable, CLI-based question answering over arbitrary document collections.

Usage

Installation

pip install raqa

Locally

pip install -e .

Run

BASH via Python interpreter

1. Build DB

   python cli.py build DATABASE_NAME PATH/TO/FOLDER/WITH/MARKDOWNS

2. Chat

   python cli.py chat DATABASE_NAME

3. One-shot retrieval

   python cli.py search DATABASE_NAME "what is retrieval augmented generation?"

4. Rebuild and chat

   python cli.py rebuild-and-chat DATABASE_NAME PATH/TO/FOLDER/WITH/MARKDOWNS

5. Get stats

   python cli.py stats

6. List databases

   python cli.py list

BASH natively

raqa build DATABASE_NAME PATH/TO/FOLDER/WITH/MARKDOWNS
raqa chat DATABASE_NAME
raqa search DATABASE_NAME "what is RAG?"
raqa list (DATABASE_NAME)
raqa stats (DATABASE_NAME)
raqa rebuild-and-chat DATABASE_NAME PATH/TO/FOLDER/WITH/MARKDOWNS

Python

Build database

from db import VectorDB
from config import MARKDOWN_ROOT

db = VectorDB()
db.build(MARKDOWN_ROOT)

Run

from agent import RAGAgent

agent = RAGAgent()
agent.chat()

Build instructions

1. If any changes are made, update pyproject.toml.
2. Building the package before uploading: cd raqa; python -m build.
3. Upload the package to pypi: python -m twine upload --repository {pypi|testpypi} dist/*

Steps 2 and 3 can be done automatically by running make publish.

Related or comparable projects

1. PyRAG difference: focuses on SingleStore.
2. ragger-simple difference: uses Qdrant and requires a Qdrant API key, in addition to an LLM API key. RAQA only requires the latter and uses an open-source tech stack for the rest.

Next steps

Real tool-calling (instead of implicit RAG)

Define OpenAI tool:

{
  "name": "search_docs",
  "description": "...",
  "parameters": { "query": "string" }
}

Hybrid search

Combine BM25 (rank-bm25) + embeddings