action is computed after flagging, using thresholds of 0.5 (review) and 0.8 (block) (or your action_thresholds):

1. If flagged is false, the action is allow.
2. Otherwise, take the highest score across all returned categories and theme results (flagged or not).
3. That score at or above 0.8 gives block, at or above 0.5 gives review, and anything lower gives allow.

Custom mode. When the request sends thresholds, action_thresholds, or the drugs_alcohol category, only flagged categories and themes count toward the highest score, and any flagged category gives at least review, even if its score is below the review level. So a category set to 1.0 can't raise the action. Requests without these fields work exactly as described above.

threshold controls flagged. The action thresholds can't be changed on this endpoint. If you need different cut-offs, use flagged and the raw score values directly.

Add "appropriateness" to categories to check whether the text fits an age rating. content_standard picks the rating (default PG-13). The result is binary: score 1 and flagged: true when the text does not meet the standard, which also makes the action block.

The appropriateness classifier can disagree with itself on borderline text. Use consensus (3 or 5) to run it several times: by default the majority wins. With consensus_mode: "unanimous", any single failing run flags the text, which is the strictest setting. Consensus doesn't change your quota usage.

themes lets you exclude subjects the standard categories don't cover, such as "gambling", "alcohol", "occult", or "heartbreak". For each theme, a classifier decides whether the text contains, references, depicts, invokes, or celebrates it, directly or through clear symbolism. It judges what the text is about, not incidental wording: "a magical evening" does not match "magic", but casting spells does.

• Input: an array of up to 10 strings, each 1 to 100 characters after trimming.
• Output: a top-level themes object keyed by your (trimmed) theme strings. Each entry has flagged, score (1 or 0), and threshold (0.5).
• Any theme match sets flagged: true and the action to block. Theme screening is meant for deliberate exclusion, not for nuanced review.
• Themes are checked alongside whatever categories you request, in the same request and for the same single quota unit.
• If the classifier returns an unreadable answer, themes are reported as not matched rather than failing the request.

Strict ("squeaky clean") screening

For a family-friendly or "squeaky clean" mode, combine a G or PG content standard, consensus: 3 (majority), and the themes your audience wants excluded:

curl -X POST https://api.moder8r.app/v1/moderate/text \
  -H "Authorization: Bearer m8r_sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Pour another round of whiskey, the night is young",
    "categories": ["toxicity", "profanity", "sexual", "violence", "appropriateness"],
    "content_standard": "G",
    "consensus": 3,
    "themes": ["alcohol", "gambling", "occult"]
  }'

{
  "request_id": "mod_8d0c6f2a-41b7-4e59-9c3d-2a7e1f5b6c04",
  "flagged": true,
  "categories": {
    "sexual": { "flagged": false, "score": 0, "threshold": 0.5 },
    "violence": { "flagged": false, "score": 0.002, "threshold": 0.5 },
    "toxicity": { "flagged": false, "score": 0.001, "threshold": 0.5 },
    "profanity": { "flagged": false, "score": 0, "threshold": 0.5 },
    "appropriateness": { "flagged": true, "score": 1, "threshold": 0.5 }
  },
  "action": "block",
  "themes": {
    "alcohol": { "flagged": true, "score": 1, "threshold": 0.5 },
    "gambling": { "flagged": false, "score": 0, "threshold": 0.5 },
    "occult": { "flagged": false, "score": 0, "threshold": 0.5 }
  },
  "sentiment": "positive",
  "usage": {
    "daily": 43,
    "monthly": 319,
    "limit": 1000,
    "remaining": 681
  }
}

Requests are checked in this order: authentication (401), request body (400), rate limit (429), monthly quota (402), then moderation. Requests rejected before moderation don't count toward your quota, and requests rejected with a 400 don't count toward your rate limit either. Only 200 responses use quota.

See the Error Reference for response shapes and retry guidance.

An older endpoint, POST /v1/moderate, is still available for existing integrations. New integrations should use /v1/moderate/text. The legacy endpoint doesn't support content_standard, themes, consensus, explain, sentiment, or the usage block. It differs in these ways:

• Body: content (required, at most 10,000 characters) plus an optional options object with categories, threshold, and action_thresholds (review, default 0.5; block, default 0.8). Unlike the text endpoint, this lets you change the action thresholds.
• Response: flagged, action, categories, and request_id.
• Errors: invalid_content, empty_content, and content_too_long (400); 402 quota_exceeded with error.usage.requests_month, error.usage.limit_month, and error.upgrade_url; 429 rate_limit_exceeded with error.retry_after (seconds); 503 moderation_rate_limit and moderation_api_error. Error bodies don't include a request_id.
• Extra headers: X-Quota-Remaining on every successful response, and X-Quota-Warning once you've used 80% or more of your monthly quota.