Tool (LLM tool use)

Last updated Jul 25, 2026

A **tool** is an external action an LLM can request — search, calculator, database query, email send — usually via function calling. The model proposes; your runtime executes (or refuses).

#When to use

When the model needs live data, precise math, or side effects it cannot do from weights alone.

#When not to

Unconstrained tools with irreversible side effects and no human approval. Never let untrusted content freely choose high-privilege tools.

#Quality checklist

Allowlist tools and arguments
Confirm destructive actions
Log every tool call
Treat tool outputs as untrusted (injection risk)

#Example

{"name": "get_order", "arguments": {"order_id": "A-1042"}}

#Learn next

Lesson: `what-are-tools`
Related: agent, MCP, function calling

Learn by doing → Learn by doing →

← All terms