Tool calling (function calling)
The model emits a structured request to run a function; the runtime executes it and returns the result.
Tool calling is how an agent affects the world. The developer declares tools with a name, description and JSON schema; the model outputs a structured call (name + arguments) instead of prose; the runtime executes it and appends the result to the context. Tools range from read-only (search, file read) to side-effecting (send email, run shell). LangChain's 2025 survey found that most teams restrict agents to read-only tools or require human approval for write/delete actions.