inference.Response is the complete result of Client.Invoke. It keeps the assistant message separate from normalized usage and terminal metadata so callers do not need to parse provider envelopes.

API surface

type Response struct {
	Message      *content.AIMessage
	Usage        *content.Usage
	Model        string
	FinishReason stream.FinishReason
	Attempts     int
}

Field	Meaning
`Message`	Assistant turn with ordered text, thinking, and tool-use blocks
`Usage`	Optional normalized token counts
`Model`	Resolved provider model name returned by the serving path
`FinishReason`	Provider-neutral terminal reason
`Attempts`	Retry establishment count; zero if uncounted, one for first-try success

response, err := client.Invoke(ctx, req)
if err != nil {
	return err
}
if response.Message != nil {
	for _, block := range response.Message.Blocks {
		if text, ok := block.(*content.TextBlock); ok {
			fmt.Println(text.Text)
		}
	}
}

A nil message or unknown finish reason is representable and should be handled explicitly. Structured-output helpers apply additional representation checks before returning data.

Proof

Source: inference/client.go
Tests: inference/client_test.go, inference/structured_result_test.go