Tool Choice Parameters
What you'll learn: How to control when and whether an LLM should invoke functions using auto, none, and required tool choice settings.
Understanding Tool Choice Control
When you give an LLM access to functions (tools), you don't always want it to have complete freedom in deciding whether to call them. Tool choice parameters let you explicitly control this behavior, similar to setting "modes" on a camera: automatic, manual, or forced.
Think of it like a restaurant menu with a sommelier. Sometimes you want the sommelier to recommend wine only if appropriate (auto), sometimes you're driving and want no wine recommendations at all (none), and sometimes you're specifically there for wine pairing and want a recommendation guaranteed (required).
The Three Tool Choice Settings
auto (Default)
The LLM decides whether calling a function is appropriate based on the conversation context. If the user asks "What's the weather in Paris?", it calls the weather function. If they ask "What's your favorite color?", it responds directly without tools.
none
The LLM will never call functions, even if they would be useful. It responds purely conversationally using its training knowledge. Use this when you want pure text responses or when tool access should be temporarily disabled.
required
The LLM must call at least one function—it cannot respond without invoking a tool. Perfect for scenarios where you always need structured action, like data extraction pipelines or when you want to force classification through a specific function.
When to Use Each Setting
auto: General-purpose agents and chatbotsnone: Explanatory responses, when external data is unavailablerequired: Guaranteed structured output, classification tasks, extraction workflows
Key Takeaway: Tool choice parameters (auto, none, required) give you explicit control over whether your LLM can, must, or cannot invoke functions, enabling predictable behavior for different use cases.