stream.FinishReason normalizes provider stop metadata without pretending every provider has the same vocabulary.

API surface

type FinishReason string

const (
	FinishReasonUnknown       FinishReason = ""
	FinishReasonStop          FinishReason = "stop"
	FinishReasonLength        FinishReason = "length"
	FinishReasonToolUse       FinishReason = "tool_use"
	FinishReasonContentFilter FinishReason = "content_filter"
)

Reason	Caller meaning
`FinishReasonUnknown`	Provider omitted or did not map a reason
`FinishReasonStop`	Normal stop; structured text can be extracted
`FinishReasonLength`	Output was cut by a limit; structured extraction fails closed
`FinishReasonToolUse`	Assistant representation is a tool call and may require continuation
`FinishReasonContentFilter`	Provider stopped for filtering; structured extraction fails closed

switch response.FinishReason {
case stream.FinishReasonStop:
	// Safe to inspect normal assistant output.
case stream.FinishReasonToolUse:
	// Continue the tool loop when the representation is a tool call.
default:
	// Handle unknown, length, and filtered output explicitly.
}

Structured output additionally rejects non-empty future reasons with StructuredOutputFinishReasonOther rather than retaining unbounded provider input.

Finish reasons

API surface

Proof