Making AI agents do real work — function calling, tool use, and building agents that take actions instead of just producing text.
Up to this point, an LLM's only output has been text — even when asked for today's weather or a calculation, it can only produce its best guess in words. Tool calling changes that: the model can decide to call a real function with structured arguments, wait for an actual result to come back, and then use that genuine result to answer you, instead of quietly hallucinating a plausible-sounding one.
