Chat moderation

Chat moderation evaluates a single chat turn against the language-model prohibition check only. The keyword blocklist is not applied, so conversational text isn’t rejected for incidental mentions of flagged terms. A turn the model judges prohibited is rejected, and the reason is returned on the completed job. Use chat moderation when you want policy-grade protection on free-form chat without false positives from a static term list. For pre-generation prompt screening that includes the blocklist, use prompt moderation instead.

Input

Submit a multipart/form-data request with:

Field	Type	Required	Description
`text`	string	Yes	The chat turn to moderate.
`webhook_url`	string	No	URL to receive the result on completion.

curl -X POST https://api.omnifence.ai/api/v1/moderate/chat \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "text=Hey, how are you doing today?"

Pipeline

Chat moderation runs a single step:

LLM prohibition check. A language model reviews the chat turn for policy violations. If it judges the turn prohibited, the job is rejected with is_prohibited: true and a short reason.

Unlike prompt moderation, the keyword blocklist scan is skipped entirely — chat moderation never short-circuits on an exact-match term.

Output

{
  "is_prohibited": true,
  "reason": "The message requests sexual content involving a minor.",
  "categories": {},
  "job_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "type": "chat",
  "status": "completed",
  "created_at": "2026-05-26T12:00:00.000Z",
  "completed_at": "2026-05-26T12:00:01.500Z"
}

When is_prohibited is true, reason explains why the chat turn was prohibited. A turn that passes returns is_prohibited: false with no reason. See decision logic for how the prohibition decision is determined.

Getting started

Platform

Moderation

Input

Pipeline

Output

​Input

​Pipeline

​Output

Input

Pipeline

Output