SqlAgent: An LLM-based Intelligent System for Natural Language to SQL Conversion

SqlAgent is an open-source natural language to SQL intelligent system built on large language models (LLMs). Its core goal is to lower the technical barrier to database querying, enabling non-technical users (such as business personnel and product managers) to query databases and obtain information using everyday language. The system has built-in multiple security verification mechanisms, supports databases like MySQL and various LLM services, and provides a practical tool for data query democratization.

Traditional SQL querying requires mastering specific syntax and understanding table structure relationships, which has an extremely high learning cost for non-technical users. With the rapid improvement of LLM's natural language processing capabilities, new possibilities have emerged to solve this problem. The SqlAgent project was born based on this trend, aiming to automatically convert everyday language into accurate SQL query statements through intelligent natural language understanding.

Natural Language to SQL Capability

Generates precise SQL code by calling LLMs using carefully designed prompt templates combined with database schema information.

Automatic Schema Loading and Context Awareness

Automatically extracts table names/comments, column information (name, type, primary key), and foreign key relationships between tables from the database's INFORMATION_SCHEMA to improve the accuracy of SQL generation.

Multiple Security Verifications

• Only allows SELECT queries; prohibits dangerous operations like DROP/TRUNCATE;
• Syntax validity check;
• Query timeout control;
• Result set size limit.

High-Performance Connection Pool

Uses DBUtils.PooledDB to implement efficient connection pool management, supports high-concurrency scenarios, and ensures stable performance.

SqlAgent uses a modular Python architecture, with core components including:

Schema Data Model and Loader

Defines database structure models, automatically extracts table structures and relationships from metadata, and is easily extensible to support new database types.

Core Agent Class (nl2sql_agent.py)

Coordinates the entire process: Schema formatting → Prompt construction → LLM calling → SQL extraction → Security verification → Query execution.

Prompt Template Design

Designed a dedicated template for MySQL (prompts/mysql.txt), which can be customized to adapt to different database types.

SqlAgent is suitable for multiple scenarios:

• Business Data Analysis: Non-technical users can directly query business data without development assistance;
• Rapid Prototype Verification: Quickly verify data query requirements in the early stages of development;
• Education and Training: Helps SQL beginners understand the correspondence between natural language and SQL;
• Internal Tool Integration: Embedded into enterprise data analysis platforms to improve efficiency.

Currently fully supports MySQL databases; future plans include expanding support to mainstream databases like PostgreSQL and SQLite. In terms of LLM services, it is compatible with all services that follow the OpenAI API format (such as OpenAI official, Tongyi Qianwen, DeepSeek, etc.), and users can choose as needed.

SqlAgent represents an innovative application of LLMs in the field of database querying, lowering the threshold for data querying and promoting data democratization (enabling non-technical users to independently access data). As LLM capabilities improve, such natural language interaction tools will become more intelligent and practical. Its open-source release provides a valuable reference implementation for this field, which is worthy of attention from developers and researchers.