Three Houses, One Idea

OpenAI shipped tool calling in mid-2023; Anthropic followed with its own version a few months later; Google added function declarations to Gemini soon after. Within a year, the same essential idea — model returns a structured tool-call object instead of trying to write JSON in prose — had three production-ready implementations.

The good news is that the concept is identical across all three. You define tools, you call the model with them, you read tool-call objects back, you execute and feed results in. If you can write the loop on one provider you can write it on the other two with a few hours of envelope translation.

The bad news is in the envelope. OpenAI calls them tools with type: "function" and a parameters field; Anthropic calls them tools with input_schema; Gemini calls them FunctionDeclarations inside a Tool object. The flow is the same; the names are not.

This track is a tour of those three envelopes. The goal is not memorization — every SDK has docs and you will look them up — but to recognize what is the same idea wearing different clothes vs. what is a real provider-specific feature you cannot port. Knowing the difference saves you from copy-pasting yourself into the wrong abstraction.

Code

Side-by-side: defining one tool, three ways·python

# OpenAI (Responses API)
openai_tools = [{
    "type": "function",
    "name": "get_weather",
    "description": "Get current weather for a city.",
    "parameters": {
        "type": "object",
        "properties": {"location": {"type": "string"}},
        "required": ["location"],
    },
}]

# Anthropic
anthropic_tools = [{
    "name": "get_weather",
    "description": "Get current weather for a city.",
    "input_schema": {
        "type": "object",
        "properties": {"location": {"type": "string"}},
        "required": ["location"],
    },
}]

# Gemini (google-genai)
from google.genai import types
gemini_tools = [types.Tool(function_declarations=[types.FunctionDeclaration(
    name="get_weather",
    description="Get current weather for a city.",
    parameters=types.Schema(
        type=types.Type.OBJECT,
        properties={"location": types.Schema(type=types.Type.STRING)},
        required=["location"],
    ),
)])]

Code

External links

Exercise

Progress

Comments 0