Natural Language Databasing. Talk to your data in English, via CLI, API or a simple web interface.

pip install nldb

An OpenAI API key should be available as the OPENAI_API_KEY environment variable, e.g. using export OPENAI_API_KEY=sk-etc.

NLDB can talk to SQLite and DuckDB databases. Call yours nldb.db or specify the name with a DATABASE environment variable.

You may need to simplify and denormalise your database for NLDB to work well. Your columns should have names whose meaning is obvious. While NLDB can join across tables, combining your data into a single table will reduce your prompt size and improve the accuracy of NLDB's query generation.

nldb init

This creates three files: index.html, Dockerfile and prompt.txt. You can ignore the first two for now.

Edit prompt.txt, following the example and instructions in the generated file. You can test the effectiveness of your prompt by pasting it into ChatGPT, replacing %s with some example questions.

nldb "What were the most watched videos by Japanese users in 2022?"

nldb serve

This starts an API server, on port 8080, with a single endpoint at /api/ask. This expects a GET request with a q parameter for the natural language query. It returns a JSON object containing the SQL statement, the result of executing the statement, a plain English answer and some timing and cost information.

There is a simple web interface to the API at /. If you don't want this, just delete index.html.

Edit prompt.txt to refine the quality of the generated queries, restarting the server to see changes. Edit index.html or replace it with your own front-end.

Generate instructions for Fly.io and Google Cloud Run

nldb deploy