REST API

POST /v1/output-guard

Moderate an assistant response with INT-Output over HTTP.

Moderate an assistant response with INT-Output before delivering it.

POST/v1/output-guard

Request#

Field	Type	Required	Description
`assistant_text`	string	Yes	The response to moderate. Must be non-empty.
`user_text`	string	No	The user message that produced it.
`messages`	array	No	Full conversation as {role, content} objects. Takes precedence over user_text.
`model_provider / model_name / session_id`	string	No	Correlation metadata.

Response#

Field	Type	Description
`label`	string	`Safe`, `Controversial`, `Unsafe`, or `Unknown`.
`severity_score`	number \| null	0.0 / 0.75 / 1.0.
`categories`	string[]	Violated categories, if any.
`refusal`	string \| null	`"Yes"` or `"No"`.
`raw_output`	string \| null	Raw moderation-model output, for debugging label parsing.
`latency_ms`	number	Server-side inference time.

Example#

curl https://integrity.triage-sec.com/v1/output-guard \
  -H "Authorization: Bearer $TRIAGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "user_text": "how do I hurt someone",
    "assistant_text": "I cannot help with that request."
  }'

JSON

{
  "label": "Safe",
  "severity_score": 0.0,
  "categories": [],
  "refusal": "Yes",
  "raw_output": "Safety: Safe\nCategories: None\nRefusal: Yes",
  "latency_ms": 455.87
}