openai-tools-decorator 1.0.2

Python module, that provides a simple way to create and manage ChatGPT tools by python decorators

openai_tools_decorator

A lightweight Python library that streamlines creating and invoking “tools” (functions) in your OpenAI ChatCompletion-based projects. It lets you register and call both synchronous and asynchronous functions via decorators.

Installation

pip install openai_tools_decorator

Quick Start

1. Import and Initialization

from openai_tools_decorator import OpenAIT

client = OpenAIT()

2. Adding Tools

Wrap your function (sync or async) with @client.add_tool(...). The decorator registers it and provides a JSON Schema describing the function’s parameters:

@client.add_tool(
    {
        "description": "Get current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {
                "city": {"type": "string", "description": "The city name in English"}
            },
            "required": ["city"]
        },
    }
)
def get_weather(city: str):
    # Or async def get_weather(...) if you prefer
    return f"Weather in {city}: 25°C"

3. Using Tools with Chat

When you call run_with_tool(...) or run_with_tool_by_thread_id(...), the ChatCompletion model can opt to invoke any matching tool. For example:

user_input = "How cold is it in Moscow right now?"
response = await client.run_with_tool(
    user_input,
    messages=[],
    model="gpt-4o"
)
print(response)  # The assistant’s response, possibly including a tool call

Example

import asyncio
import aiohttp
from openai_tools_decorator import OpenAIT

client = OpenAIT()
api_key = "<YOUR_API_KEY>"

async def fetch_url(url):
    async with aiohttp.ClientSession() as session:
        async with session.get(url) as resp:
            return await resp.text()

@client.add_tool(
    {
        "description": "Fetch weather from an API",
        "parameters": {
            "type": "object",
            "properties": {
                "city": {
                    "type": "string",
                    "description": "The city name in English"
                }
            },
            "required": ["city"]
        },
    }
)
async def get_weather(city: str):
    url = f"https://api.openweathermap.org/data/2.5/weather?q={city}&appid={api_key}&units=metric"
    return await fetch_url(url)

async def main():
    question = "What's the temperature in London?"
    result = await client.run_with_tool(question, messages=[])
    print("Assistant says:", result)

asyncio.run(main())

Key Points

• You can decorate both synchronous and asynchronous functions.
• Tools get automatically registered and described for the OpenAI model.
• The model decides whether or not to call your tool during the conversation.

License

Distributed under MIT or any other license of your choice. Contributions and feedback are always welcome!