Question 1

What is function/tool calling?

Accepted Answer

It's the protocol by which an LLM can request that your code execute a function. You declare the functions it can call (name, description, JSON Schema of parameters); the model decides when to call them and returns a structured `tool_calls` array; you execute and return results; the loop continues. This is how agents do anything beyond text, query a database, call an API, run a calculation, control a UI.

Question 2

What's the difference between OpenAI's tools and the older functions API?

Accepted Answer

Same schema, slightly different wrapper. The old API used `functions: [...]` and the response had `function_call`. The current `tools: [{type: 'function', function: {...}}]` supports multiple tool types (function, file_search, web_search, code_interpreter) and the response has `tool_calls: [...]` which can include parallel calls. The schemas inside the `function` object are identical between both APIs.

Question 3

Does the description field matter?

Accepted Answer

Critically. The model sees the function name, description, and parameter descriptions, that's how it decides when to call. "get_weather" with description "Get current weather for a city" gets called constantly when users ask about weather. With no description, it gets ignored even when relevant. Same applies to each parameter: "city" with description "City name, e.g. 'Paris' or 'New York, NY'" produces better arguments than just "city".

Question 4

Which JSON Schema features are supported?

Accepted Answer

OpenAI supports a subset: `type` (string, number, integer, boolean, array, object, null), `properties`, `required`, `enum`, `items` (for arrays), `description`. With structured outputs (`strict: true`), additional constraints: `additionalProperties: false` is mandatory, all properties must be in `required`, no `oneOf`/`anyOf` at root, no `pattern`, no `format` except specific ones. The strict subset is narrower but guarantees the model produces valid JSON.

Question 5

What's the structured outputs / strict mode?

Accepted Answer

Setting `strict: true` on a function makes the model output guaranteed to validate against the schema, OpenAI uses constrained decoding so the LLM literally can't emit invalid JSON for that function. Caveats: stricter schema requirements (all fields required, no optionals), every property must explicitly declare `additionalProperties: false`, and the schema is compiled on first use (latency on first call). Use strict for production; non-strict for prototyping.

Question 6

How many tools can I define?

Accepted Answer

Soft limit ~128, beyond that, context bloat from tool descriptions eats your prompt budget and the model gets worse at choosing. Hard rule: each tool description and parameter description counts against your token limit. 50 tools with 100-token descriptions = 5K tokens consumed before the user's prompt. Group related tools, or use a router pattern: a single 'dispatch' tool that calls sub-tools server-side.

Question 7

Should function names be camelCase or snake_case?

Accepted Answer

Snake_case is the convention in OpenAI's docs and examples, and what most agent frameworks (LangChain, OpenAI Agents SDK) use. The model handles either fine, it's a convention, not a requirement. Pick one and be consistent across your tool set. Mixed conventions confuse no one but the humans reading your code.

Question 8

Is my schema sent to OpenAI when I'm building it?

Accepted Answer

No, schema building is local. The schema only travels to OpenAI when you actually make an API call with it as part of `tools`. Schemas themselves aren't sensitive in most cases, but their structure can reveal your internal API surface, which some teams treat as confidential.

Question 9

What standards does OpenAI Function Schema Builder validate against?

Accepted Answer

OpenAI Function Schema Builder checks compliance with Output format specification.

Question 10

What are the limitations of OpenAI Function Schema Builder?

Accepted Answer

Output is template-based, may need customization Does not validate against target system

JSON Feature	Standard Mode	Strict Mode Requirements
`additionalProperties`	Optional (Default True)	Must explicitly be `false` on all objects
`required` Fields	Optional properties allowed	All defined properties must be in the required array
Nullable Values	Implicit via omission	Must use `["type", "null"]` array syntax
Polymorphism	`anyOf` / `oneOf` allowed anywhere	Restricted to property-level definitions only

OpenAI Function Schema Builder

Parameters

Schema JSON

Integration Guide

Learn More

Configuration Mastery: Taming the YAML vs. JSON vs. TOML War

We Benchmarked 6 JavaScript JSON Parsing Approaches: What Actually Wins in Practice

JSON Schema Validation: Stop Trusting API Responses Blindly

What is OpenAI Function Schema Builder?

Frequently Asked Questions

Technical Deep Dive

OpenAI Function Schema Builder

AI-Augmented

Practical Output

Privacy-First

01 Schema Strict Mode Constraints

02 Agent Execution Pipeline

Runs 100% in your browser

You Might Also Need

AI Prompt Cost Calculator

AI Code Explainer

LLM Model Comparison Reference

Structured Output Schema Builder