RaiInvocationModeration interface

Declares where request/response text lives so content-safety guardrails can extract it.

Properties

input_content_type	How the REQUEST body is parsed. When omitted, the service defaults to `json`.
input_paths	Path(s) to user text in the REQUEST body. Required when input_content_type is `json`.
output_content_type	How the RESPONSE body is parsed. When omitted, the service defaults to `json`.
output_paths	Path(s) to agent text in a NON-STREAMING response body. Required when response_mode is non_streaming/both and output_content_type is `json`.
response_mode	Author-declared response shape; drives which output gate runs and which fields are required.
stream_selectors	One SSE event->field selector per event type carrying text. Required when response_mode is streaming/both and output_content_type is `json`.

How the REQUEST body is parsed. When omitted, the service defaults to json.

input_content_type?: RaiInvocationContentType

Path(s) to user text in the REQUEST body. Required when input_content_type is json.

input_paths?: string[]

string[]

How the RESPONSE body is parsed. When omitted, the service defaults to json.

output_content_type?: RaiInvocationContentType

Path(s) to agent text in a NON-STREAMING response body. Required when response_mode is non_streaming/both and output_content_type is json.

output_paths?: string[]

string[]

Author-declared response shape; drives which output gate runs and which fields are required.

response_mode: RaiInvocationMode

One SSE event->field selector per event type carrying text. Required when response_mode is streaming/both and output_content_type is json.

stream_selectors?: RaiSseTextSelector[]