A simple, multi-user, multi-conversation, web-based chatbot to answer questions based on database queries.
This project is evolved from https://github.com/edwardzjl/chatbot, utilizing langchain's sql agent and offers the following features:
- Enhanced Table Descriptions: This feature provides additional table descriptions, empowering Language Model (LLM) to make more informed decisions when selecting target tables.
- Customizable Data Definition Language (DDL, the
CREATE TABLE
statement): Typically, SQL agents are not able to modify the DDL of the database. However, the original DDL often contains extraneous information. Moreover, at least in Postgres, the DDL lacks the column descriptions, which are crucial for LLMs to accurately choose the appropriate columns. - Redis Integration: Both the supplementary table descriptions and customized DDL are conveniently stored in Redis. This approach ensures swift access and facilitates straightforward updates.
A live demo was served at https://sqlbot.agi.zjuici.com
This demo was built on IMDb dataset, with 6 tables.
Key | Default Value | Description |
---|---|---|
LOG_LEVEL | INFO |
log level |
REDIS_OM_URL | redis:https://localhost:6379 |
Redis url to persist messages and metadata |
ISVC_LLM | https://localhost:8080 |
model service url |
WAREHOUSE_URL | postgresql+psycopg:https://postgres:postgres@localhost:5432/ |
data warehouse url, in which you want to analyze the data |