OpenAI Function Calling: Building Reliable Tool-Using Agents

What Is Function Calling?

OpenAI's function calling (now called tool use in the API) lets the model output structured JSON matching a schema you define — instead of free-form text. This makes calling real code from LLM outputs trivial without fragile regex parsing.

Defining Tools

tools = [{
  "type": "function",
  "function": {
    "name": "get_weather",
    "description": "Get current weather for a city",
    "parameters": {
      "type": "object",
      "properties": {
        "city": {"type": "string", "description": "City name"},
        "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
      },
      "required": ["city"]
    }
  }
}]

The Complete Agentic Loop

import openai, json

def run_agent(user_message):
    messages = [{"role": "user", "content": user_message}]
    while True:
        resp = openai.chat.completions.create(
            model="gpt-4o", messages=messages,
            tools=tools, tool_choice="auto"
        )
        msg = resp.choices[0].message
        messages.append(msg)
        if msg.tool_calls:
            for tc in msg.tool_calls:
                result = dispatch_tool(tc.function.name, json.loads(tc.function.arguments))
                messages.append({"role":"tool","tool_call_id":tc.id,"content":json.dumps(result)})
        else:
            return msg.content  # Final answer

Parallel Tool Calls

GPT-4o can call multiple tools in a single turn — drastically reducing latency for tasks that need several pieces of information simultaneously:

Parallel tool calling can reduce agent latency by 50–70% for tasks requiring multiple independent lookups.

Structured Outputs with Strict Mode

Setting "strict": true in your tool definition guarantees the model's output exactly matches your schema — no extra fields, no missing required properties. Essential for production agents where downstream code depends on the output shape.