What is a tool?

A tool is a function the model can ask your code to run. Tools are how text output becomes real action: querying a database, calling an API, reading a file, or sending a message.

Registering a tool is declarative work. You provide a name, a description, and a JSON Schema for the parameters, and pass this definition along with the prompt. The model never executes anything itself. When a reply would benefit from outside data or an effect, it returns a structured request: a tool name plus arguments it composed from the conversation. Your code validates those arguments against the schema, runs the actual function, and sends the result back into the message list. The model then continues with the fresh data in context. Both major APIs formalize this handshake: OpenAI documents it as function calling, Anthropic as tool use.

Well-designed tools share the properties of good APIs: small surfaces, precise descriptions, and typed arguments, because the model chooses between them based mostly on the descriptions you wrote. Tools also set the trust boundary. The schema decides what the model may request, and your code decides what actually runs.