This is a barebones Python demo that showcases how to build a simple conversational agent (in terminal) using OpenAI's API with tool calling. The focus is on the core logic, clarity and simplicity — no external frameworks or complex abstractions.
🔧 Just Python + OpenAI API — nothing more.
This example uses basic math tools (add, subtract, multiply, divide) to demonstrate the key components of tool calling:
- Defining tools as functions
- Detecting when the model wants to use them
- Executing the function with provided arguments
- Feeding the result back into the conversation for further reasoning
This code is intended as a simple, functional starting point, and easy to plug your own tools or logic.
- Install requirements using the command:
pip install -r requirements.txt - Make sure the OpenAI API key is in the environment. You can do
export OPENAI_API_KEY=<YOUR_KEY>from the terminal. - Then, run the command:
python run_chat_script.py.
-
Auto-generating Tool Definitions from Python Functions:
OpenAI's tool calling requires every tool to be defined in a JSON schema — specifying parameters, types, and descriptions. Manually writing these schemas becomes tedious, especially if you're working with multiple or complex functions.
To keep things simple, this project includes a helper function: func_to_json (source: funkagent). It uses Python type hints and docstrings to auto-generate valid tool definitions — reducing boilerplate and making it easy to scale.
There are other tools like Lunary's tool call generator as well though I haven't tested their support for more complex use cases.
-
Parallel Tool Calling (Optional)
By default, the script enables parallel tool calling, allowing the model to call multiple tools in a single turn if needed. However, in this script, it's disabled to let the model solve math expressions step by step — for example, evaluating "555 * 222 - 121" in sequence. To enable it, you can set
parallel_tool_callstoTrue.response = client.chat.completions.create( ... parallel_tool_calls=True )