A developer is building a Foundry-hosted travel assistant that needs to check real seat availability from the airline's reservation system before confirming a booking to a customer. Instead of asking the model to guess or describe availability in prose, which approach should the developer use so the model reliably invokes the airline's API with correctly structured parameters?

The correct approach is defining a function or tool with a strict parameter schema and letting the Foundry-hosted model call it when needed. That lets the model recognize when a request requires live seat data and invoke the airline's API with correctly typed, structured parameters instead of generating an answer from its own text patterns. The system-message approach is incorrect because a system message can only set static instructions and phrasing rules; it cannot connect to a live reservation system or guarantee the model produces machine-readable, correctly structured request parameters. The fine-tuning approach is incorrect because fine-tuning teaches the model to reproduce patterns from past examples, which means it would generate plausible-sounding but potentially stale or fabricated availability rather than checking the live system at booking time. The higher-temperature approach is incorrect because raising temperature increases the randomness and variety of generated text, which works against the goal of reliably invoking an external system with precise, consistent parameters, and can make outputs less predictable rather than more accurate. Since the travel assistant must retrieve real, current seat availability and act on it correctly, structured function calling is the mechanism designed specifically to let a model trigger an external system with validated inputs rather than describe an answer in prose.