In this tutorial, Julia Muiruri illustrates how developers can leverage GPT-5’s FreeForm tool calling within Azure AI Foundry to execute SQL and JavaScript workflows for advanced analytic tasks, highlighting agentic automation and custom tool design.

GPT-5 FreeForm Tool Calling in TypeScript on Azure AI Foundry

Julia Muiruri demonstrates a practical, code-heavy workflow where a Data Analyst automates species-level mean calculations and visualization for the Iris dataset using a GPT-5 agent.

Scenario Overview

Carl, a Data Analyst, avoids manual SQL and chart logic. Instead, he prompts an internal GPT-5 Agent: “Compute species-level means, then visualize petal length.”

  • Custom Tool 1: sql_exec_csv – Runs raw SQL on an Iris dataset, returns CSV.
  • Custom Tool 2: code_exec_javascript – Executes raw JavaScript in a secure VM, parses CSV, generates and saves an SVG chart.

GPT-5 orchestrates multi-stage execution:

  1. Generates SQL, receives CSV results.
  2. Emits JavaScript code for table and chart display.
  3. Chains these steps without JSON schema wrapping.

What Is FreeForm Tool Calling?

FreeForm tool calling allows GPT-5 to invoke registered tools with arbitrary, unstructured text payloads—SQL, JavaScript, Python scripts, Bash, and more—avoiding JSON argument schemas.

Key Benefits

  • No Schema Friction: Use the tool’s native language (SQL, JS), skip JSON wrapping/parsing.
  • Improved Reasoning: Mixed output (code, natural language) facilitates better AI workflows.
  • Multi-Step Chaining: Each tool’s output re-enters the workflow as plain text for further model reasoning.

Key Differences

Dimension Structured Function (JSON Schema) FreeForm Custom Tools
Payload Shape JSON arguments Raw text (code/script)
Validation Automatic via schema Semantic, at execution
Parsing Overhead JSON mapping Minimal (direct string)
Ease of Evolution Update schema/code Update prompt/tool description
Observability Nested JSON hides code Natural language/code visible
Chaining Complexity Must parse JSON per step Direct string passing
Errors JSON structure errors Runtime execution errors

When Not To Use FreeForm

  • Strict validation (financial, PII)
  • Complex, deeply nested data
  • Mass extraction (needs consistent JSON for parsing)

Hybrid designs can mix structured and custom tools for validation and execution tasks.

Implementation Breakdown

Prerequisites

  • Azure AI Foundry project
  • GPT-5 deployment (v1 API)

User Workflow

  1. Prompt GPT-5 to write SQL (means on Iris species) for CSV output.
  2. Prompt to generate JavaScript for parsing the CSV, printing a table, and rendering a bar chart (mean petal length by species).

Tool Registry:

  • sql_exec_csv: Executes SQL, returns CSV string.
  • code_exec_javascript: Executes JavaScript, produces chart and console table.

Execution Flow:

  • Model emits raw SQL (not JSON).
  • Tool executes query, returns CSV.
  • Model emits JavaScript (not schema-wrapped).
  • Tool executes code, prints table, saves SVG chart.

Benefits for Developers:

  • Eliminates friction of JSON parsing; tools work with direct code.
  • Enables creative, expressive, and rapid prototyping using agentic patterns.
  • Azure AI Foundry supports governance and scaling for production deployments.

Reference Links & Resources

Author: Julia Muiruri

Published: Nov 14, 2025

Platform: Microsoft Developer Community Blog

This post appeared first on “Microsoft Tech Community”. Read the entire article here