minimal-db-agent

A minimal Python agent that uses LLMs to answer questions about an unknown SQLite database at inference time, leveraging ReAct-style reasoning and SQL queries.

Features

• Uses OpenRouter-hosted LLMs via LangChain for natural language understanding.
• Profiles and queries any SQLite database without prior schema knowledge.
• Implements a ReAct agent with tool support for SQL querying.
• Returns both natural language answers and the SQL used to derive them.
• Designed for extensibility and clarity.

Structure

• core/
  • agent.py: ReAct agent creation and invocation logic.
  • model.py: LLM wrapper and model loader.
  • tools.py: SQL query tool for database interaction.
  • system_prompt.txt: Agent instructions and behavior specification.
• data/db.sqlite: Store your SQLite database.
• demo.py: Example script to run the agent on a sample question.
• secrets/.env: Stores API keys (e.g., OPENROUTER_API_KEY).

Installation

1. Python 3.11+ required
2. Install dependencies:

   pip install -r requirements.txt

3. Add your OpenRouter API key to secrets/.env:

   OPENROUTER_API_KEY=your-key-here
   

Usage

Run the demo script:

python demo.py

The agent will connect to data/db.sqlite, profile the database, and answer the question:
“What is the database about?”

Customization

• Replace data/db.sqlite with your own SQLite database.
• Modify core/system_prompt.txt to change agent behavior.
• Extend core/tools.py to add more tools.

Requirements

• Python 3.11+
• langchain-openai, langgraph, sqlalchemy, dotenv