Agentic AI Notebook
Tool Calling & Function Calling
Phase 7Module 14 of 15

SQL Tool

Warehouse truth beats hallucinated numbers. An open SQL prompt also drops tables.

A teller window into the vault — not handing over the master key.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / to zoom up to 200% if needed.

Overview

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Ask then query

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Model fills parameters. You run a template.

Key Takeaways

  • 1.A SQL tool lets the agent query data you already trust. Prefer parameterized, read-only queries on a view.
  • 2.Never concatenate model text into SQL. Cap rows and time; return a compact table, not 100k rows.
  • 3.Templates or a schema-limited query builder. DB user is SELECT-only.
  • 4.Bind parameters. Summarize wide results.

Learn elsewhere

  • Tool Validation
  • External APIs

Real Example

Scenario

lookup_order(id) runs SELECT … WHERE id = $1 LIMIT 1. The model never sees a string it can turn into OR 1=1.

What you would do

In Tool Calling & Function Calling, apply SQL Tool to this scenario: lookup_order(id) runs SELECT … WHERE id = $1 LIMIT 1. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Commands

Commands to Remember

  • Read-only role
  • Bound parameters
  • LIMIT + timeout
  • No concat

Cheat Sheet

Quick recap

quick ref
  • Teller window
  • Templates > free SQL
  • Cap rows
  • Writes need HITL

Common Mistakes

  • f"SELECT * FROM {table}"
  • Admin DB credentials on the agent
  • Returning 50k rows into the prompt