What is tool calling?
Tool calling — also called function calling — is the mechanism by which a language model asks for a named function to be executed with structured arguments. The model does not run anything itself; it emits a request, and your code decides whether to honour it.
This distinction is load-bearing and widely misunderstood. A tool call is a suggestion in a structured format. Everything about whether it happens, with what credentials, against which tenant, and whether a human sees it first, is decided by the layer that receives it.
Which means "the model called the wrong tool" is rarely the real failure. The real failure is that the wrong tool was reachable.
Actions become callable only after a person reviews them during import and publishes them. Anything not published cannot be called, whatever the model emits.
Related terms
- AI agent — A system that uses a language model to decide which actions to take, then takes them — as opposed to only producing text.
- MCP (Model Context Protocol) — An open protocol for exposing tools, data and prompts to language models through a consistent interface.
- OpenAPI — A machine-readable specification describing a REST API: its endpoints, parameters, schemas and responses.
See it working rather than described.
The product demo runs one agent across three systems and stops at the approval gate.