Tool calling, also called function calling, is the mechanism that lets a large language model invoke external functions, APIs, or systems to take real actions rather than only generating text. The model decides which tool to call and with what arguments, the tool runs, and the result is fed back into the model's reasoning.
It is the bridge from talking to doing. Without tool calling, a model can describe how to issue a refund or look up an order. With tool calling, it can actually query the order system, draft the refund, and update the ticket. This is the core capability behind most agentic AI for customer experience.
Tool calling is a model-layer capability, not an Aide-owned term. In Aide, the agentic AI platform for customer experience, tools are not exposed to the model freely. They are scoped inside ASOPs (Agentic SOPs), intent-scoped and test-gated, so a given automation can only call the tools its verified procedure allows.
The model reaches only the tools its verified procedure grants it, and every call is recorded and reviewable, so an automation cannot take an action that never cleared review or hide one it took. Each tool-using procedure belongs to a named customer intent, which keeps the logic the team's to read, revise, and learn from, not an opaque agent's to keep.