Language models & chatbots
Tool calling lets a language model request that an outside program run, such as a calculator, database, or web search, and then use the result in its answer.
A language model can only produce text. It cannot check today's weather, do exact arithmetic reliably, or send an email. Tool calling fixes that. The developer describes the available tools to the model. When the model decides one is needed, it writes a structured request naming the tool and the inputs. The app runs it and hands back the result.
The model never runs anything itself. It writes the order; the surrounding software cooks it. That separation matters for safety, because the app decides which tools exist and can refuse or double-check risky requests. The model then continues its reply using whatever came back.
Tool calling is what turns a chatbot into an agent. It is also how assistants read your calendar, look up live prices, or query a company database. Standards like the Model Context Protocol, introduced in 2024, make it easier to connect many tools to many models.
Asked 'what is 17 percent of 3,482,' the model calls a calculator tool, receives 591.94, and reports it rather than guessing the arithmetic.