LLM APIs and Tool Calling

What model APIs provide

• Chat completions with role-structured messages.
• Model choices based on cost/latency/quality.
• Vision/audio/embeddings support in selected models.
• Production-grade usage with token-based billing.

Tool calling concept

Model converts user intent into a structured function call (JSON arguments), then app executes the function and returns result back to model.

1. User asks a question.
2. Model selects tool and fills arguments.
3. App executes tool (DB/API/calculation).
4. Result is fed back for final natural-language answer.

Why tool calling matters

• Connects LLM reasoning with real-world systems.
• Reduces hallucination by grounding on tool outputs.
• Enables agent workflows (search, calculate, transact).

Basic schema pattern

{
  "name": "get_weather",
  "parameters": {
    "type": "object",
    "properties": { "location": { "type": "string" } },
    "required": ["location"]
  }
}