DBChat

An agentic AI tool designed to seamlessly interact with existing databases using an agentic AI approach. This project aims to simplify the process of querying databases, making it more intuitive and accessible for everyone, regardless of their technical background.

Features

• Uses an Agentic approach to answer to user's question
• Safe from user executing malicious DDL queries.
• Can add Data-safety by only exposing certain tables and columns in the schema.
• Doesn't expect the LLM to generate executable code, rather behaves as an reasoning engine
• reduced hallucinations and more deterministic in nature
• no need for vector DBs
• Supports all relational databases supported by sql alchemy. Refer https://docs.sqlalchemy.org/en/20/dialects/

How to Run?

docker pull dhanilan/dbchat
docker run -p 5173:5173 -p 8000:8000 dhanilan/dbchat

default database to save connections and conversation is mongo at mongodb://localhost:27017/dbchat

if you want to save them at specific mongo database pass it as an env

docker run  -p 5173:5173 -p 8000:8000 -e DB_URL=mongodb://host.docker.internal:27017/dbchat dhanilan/dbchat

if you want to run spider dataset from https://yale-lily.github.io/spider

docker run  -p 5173:5173 -p 8000:8000 -e DB_URL=mongodb://host.docker.internal:27017/dbchat -e ATTACH_SPIDER_DATASET=1 dhanilan/dbchat

using docker-compose

The docker compose also comes with a chinook database for testing. You can use it by adding a connection to the chinook database in the UI use postgresql+psycopg2://postgres:postgres@chinook:5143/chinook as the connection string

update the docker-compose.yml with the required envs if neccessary and run

docker-compose up -d

Architecture

Built with Autogen.

Development

Dev container contains all the necessary deps , mongodb storing the schema and chat history , conversations, settings etc and also a chinook database in postgres for playing around

Installing dependencies

UI

pip install -r requirements.txt

server

cd src/ui && npm i

To run the Application in local

UI

cd src/ui && npm run dev

API

cd src && uvicorn api.main:app

Open the UI

By default opens in http://localhost:5173/

Goto setting to save the OPEN AI API key

connect to dev container chinook db

postgresql+psycopg2://postgres:postgres@localhost:5432/chinook

Roadmap

• clean up some code for now
• Add more tests
• add feature to annotate the schema
• add few shots of the conversation to the schema
• add feature to add custom prompts
• add access control for tables and columns `