Glossary
tool use
The general pattern of an LLM invoking external functions, APIs, or systems to fetch data or take action, distinct from generating an answer purely from its weights.
A broader umbrella than function calling. function callingagentsA pattern where a model emits a structured call (function name plus arguments), the runtime executes it, and the result returns as input on the model's next turn.
Open full entry is the
specific structured protocol; tool use is the capability and the
practice. A model “uses tools” when it decides to call a function,
reads the result, and integrates it into its response. Whether the
implementation is a JSON tool-call schema, a code-interpreter sandbox,
a structured plan, or an inline shell command depends on the agent
framework.
The dimension models compete on is decision quality. Knowing when to call a tool versus when to answer from prior knowledge is harder than the call itself. Calling the right tool from a large catalog is harder than calling the only available tool. Recovering from a tool error gracefully is harder than success. Benchmarks like τ-bench and SWE-Bench measure these end-to-end agent behaviors directly.
Tool use is what separates a chat model from a production agent (coding agents, customer-support agents, computer-use agents). It is also the primary attack surface: prompt-injected tool calls are the dominant agent-security failure mode.