Structured Tool Use
Making the model’s tool requests conform strictly to a defined shape, so they can’t come out malformed.
Think of it like
A form that won’t submit until every required field is filled with the right kind of value — no half-finished orders slip through.
Example
The model must return `{amount: number, currency: string}`; the system rejects or repairs any call missing a field or with the wrong type.
How it actually works
Rather than hoping the model emits valid arguments, structured tool use constrains generation to a schema — via grammars, constrained decoding, or validate-and-retry — so the output is guaranteed parseable. This removes a whole category of flaky failures where a stray character broke the JSON. The tradeoff is a little rigidity and setup, in exchange for calls your code can trust.
For product teams
The unglamorous reliability layer that keeps agents from breaking on a misplaced comma.
For engineers
Enforcing schema conformance on tool-call arguments via constrained decoding or validation-and-repair.
Related
- Tool Schema — The schema it enforces against.
- Tool Calling — The call it makes trustworthy.
- Function Calling — The structured-output mechanism it tightens.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome