Forcing and Disabling Function Calls
You'll learn how to control whether the LLM must call a function, can choose freely, or cannot call functions at all.
The Core Idea
When working with function calling, you don't always want the LLM to decide on its own. Sometimes you need to force it to call a specific function (like always using a calculator for math). Other times you want to disable function calling entirely (like when you just want a conversational response). The tool_choice parameter gives you this control.
Three Control Modes
Think of tool_choice like setting your car's transmission:
1. Automatic (default)
tool_choice="auto" # LLM decides whether to call functions
The LLM chooses whether to call any function based on the user's request.
2. Manual/Forced
tool_choice={"type": "function", "function": {"name": "get_weather"}}
Forces the LLM to call the specific function you name. Use this when you know a function is required.
3. Disabled
tool_choice="none" # No function calls allowed
Prevents all function calling. The LLM responds conversationally even if functions are defined.
When to Use Each
auto: Normal chatbot interactions where context determines if tools are needed- Forced: Multi-step workflows where each step requires a specific tool
none: Getting explanations, summaries, or conversational responses without triggering tools
Key Takeaway: The tool_choice parameter lets you override the LLM's decision-making—forcing specific function calls when certainty is needed, or disabling them when only text responses are appropriate.