Quickstart

Open your terminal. We’re going to create a folder called my_agent, set up a virtual environment, and install the minimal dependencies.

mkdir my_agent
cd my_agent
uv venv
uv pip install opentools-sdk openai python-dotenv

OpenTools reads credentials from environment variables. For this quickstart we’ll use Alpaca because of its strong capabilities and easy setup.

​

Model API key

• Navigate to the OpenAI dashboard (API keys section)
• Click Create new secret key
• Copy the key and store it somewhere safe (OpenAI recommends a secure place like your .zshrc but ensure to add it to your .env file)

​

Alpaca API key

• Create an Alpaca account (no funding required for paper trading)
• In the dashboard homepage, find Your API Keys
• Click Generate New Keys
• Copy:
  • Key ID
  • Secret Key (shown only once)

​

Create a .env file

This quickstart uses paper trading by default and never places live trades unless you explicitly enable it. OpenTools does not store keys or manage OAuth. All credentials stay on your machine and are only used when tools are called.

In the same folder as main.py, create a file named .env:

OPENAI_API_KEY="your_openai_api_key"
ALPACA_KEY="your_alpaca_key_id"
ALPACA_SECRET="your_alpaca_secret_key"

You’re about to run a small tool loop: the LLM decides which trading tool to call, OpenTools executes it, and the LLM summarises the result.

​

Create a main.py

Create a file named main.py and paste the following code:

import asyncio
import os

from dotenv import load_dotenv
from openai import AsyncOpenAI

from opentools import trading
from opentools.adapters.models.openai import run_with_tools


async def main() -> None:
    # load OPENAI_API_KEY, ALPACA_KEY, ALPACA_SECRET from .env
    load_dotenv()

    # this quickstart uses no framework, just the model
    client = AsyncOpenAI(api_key=os.environ["OPENAI_API_KEY"])

    # add the alpaca tools
    service = trading.alpaca(
        api_key=os.environ["ALPACA_KEY"],
        api_secret=os.environ["ALPACA_SECRET"],
        # specify the model, this is required for bundling tools correctly
        model="openai",
        # specify paper trading account or live trading account
        paper=True,
        # specify minimal output returned to model (maximum token efficiency)
        minimal=True,
    )

    prompt = "Show my account summary and list any open positions."

    # run_with_tools creates the tool-loop to make your development time quicker
    result = await run_with_tools(
        client=client,
        model="gpt-4.1-mini",
        service=service,
        user_prompt=prompt,
    )

    print(result)


if __name__ == "__main__":
    asyncio.run(main())

The code above will add the tools automatically to the LLM context and use a loop to relay tool responses. The tools output structured schemas to ensure interpretability is maximised and relay the information accurately - this has been tested with LLMs as small as 7 billion parameters with clear, correct results.

​

Run your code

uv run python main.py

If the model responds without calling tools, ensure you structure yours prompts directly around the tool, here are some examples: “Get my account”, “List positions”, or “Show recent orders”.