toolsmith

Toolsmith turns your Python functions into structured, AI-callable tools, using type hints and docstrings. Integrates with any LLM provider that is compatible with the OpenAI API.

• Very easy to use: Just write normal Python functions with typehints.
• Pydantic support: Pydantic types get automatically serialized and deserialized.
• Unopinionated: Toolsmith gets out of the way in terms of how you want to wire up the LLM loop.
• Fast: Highly performant based on Pydantic's speed.
• Async support: Built-in support for async functions with parallel execution of multiple tool calls.

Installation

$ pip install toolsmith

Usage

Simply define any functions you may have normally:

def create_user(name: str, age: int) -> str:
    """Saves a user to the DB"""
    return f"Created user {name}, age {age}"

def search_users(query: str, regex: bool) -> str:
    return "Found some users"

Put it all together and call the OpenAI API:

import openai
from toolsmith import AsyncToolbox

toolbox = AsyncToolbox.create([create_user, search_users])

client = openai.AsyncOpenAI()
response = await client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Make a 33 year old user called Alice"}]
    tools=toolbox.get_schema(),  # toolsmith handles the schema
)

# toolsmith will automatically call your functions here
results = await toolbox.execute_tool_calls(response.choices[0].message.tool_calls)

Learn more

https://dougli.github.io/toolsmith/