sqlchat

Chat with an SQLite auto-parts inventory in plain English. A LangChain agent inspects the schema, writes SQL, runs it read-only and explains the result. Follow-ups like "now only the ones under ₹500" work because the conversation is stored per thread in chat_memory.db, so it also survives restarts.

Setup (Windows PowerShell)

python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
copy .env.example .env      # then paste your ANTHROPIC_API_KEY into .env
python seed_db.py           # builds inventory.db with sample data
python main.py

Using it

Commands: /new starts a fresh conversation, /help shows tips, /quit exits. The bot prints a [tool] line each time the agent calls a tool, so you can watch the loop: list tables, describe, query, answer.

Try:

• Which brake parts are out of stock?
• Show Hyundai Creta parts under ₹3000
• Now only the ones from Bosch India
• Which supplier has the most stock value?

Files

File	Purpose
seed_db.py	Builds inventory.db (suppliers, vehicles, parts, fitment, stock)
db.py	Read-only connection, SELECT-only guard, 50-row cap
tools.py	list_tables, describe_table, run_query
agent.py	Model, system prompt, SQLite checkpointer, agent graph
main.py	CLI loop with streaming and /new, /help, /quit

Safety

The database opens in read-only mode, only a single SELECT or WITH query passes validation, and results are capped at 50 rows. Errors go back to the agent as text so it can fix its SQL and retry.

Your own database

Add DB_PATH=C:\path\to\your.db to .env. The agent discovers tables and columns itself; you may want to adjust the prices and currency line in the system prompt in agent.py.

Troubleshooting

• greenlet fails to build: use Python 3.12 (for example conda create -n sqlchat python=3.12), or run pip install --only-binary=:all: --upgrade greenlet first.
• ImportError on create_agent: harmless; agent.py falls back to LangGraph's create_react_agent automatically.
• Weird characters instead of ₹: run chcp 65001 in the console first.