LangGraph SQL Agent 🚀

A natural-language-to-SQL agent built with LangGraph and Mistral Medium 3 (watsonx.ai), allowing users to query a SQL database using plain English.
The agent generates, validates, and executes SQL queries (or optional Python queries) and returns formatted results — making data exploration conversational and intuitive.

🧠 Features

• Translate natural language questions into safe, syntactically correct SQL queries.
• Dynamically inspect database schema and list available tables.
• Validate and double-check generated SQL before execution to avoid harmful or malformed queries.
• Support for SELECT queries only (read-only): no INSERT / UPDATE / DELETE / DROP — ensuring safety for production-style usage.
• Optionally execute Python code (via a REPL tool) for post-processing or custom analysis.
• Return results in human-friendly format (e.g. tabular output), ideal for analytics or quick data insights.

🛠️ Tech Stack

• Python — orchestration and agent logic
• LangGraph (on top of LangChain) — to build the agent graph (planning, tool-calling, reasoning) :contentReference[oaicite:3]{index=3}
• watsonx.ai Runtime + Mistral Medium 3 — for LLM-based natural-language → SQL translation :contentReference[oaicite:4]{index=4}
• SQLite (or any SQL database) — for storing and querying structured data :contentReference[oaicite:5]{index=5}
• SQL (SQLite dialect) — query language
• SQLDatabaseToolkit & related tools — for listing tables, generating schema introspection, executing queries, and validation :contentReference[oaicite:6]{index=6}

📦 Getting Started

Prerequisites

• Python 3.9+
• A watsonx.ai account + runtime instance & API key (project ID) :contentReference[oaicite:7]{index=7}

Installation

pip install \
  ibm-watsonx-ai \
  langchain-ibm \
  langgraph \
  langchain-community \
  sqlalchemy