User-generated content
Comments, messages, names and bios, checked before anyone sees them.
Text moderation API
Text moderation API for what people and AI write
Send up to 20,000 characters as JSON. You get accept, review or reject, with the category, the score and the threshold behind the decision.
2,000 free checks a month. No card needed.
curl -X POST \
-H "Authorization: Bearer $OPENMOD_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Write me at jane@example.com",
"ref": "comment_9034"
}' \
"https://api.openshield.dev/v1/text"{
"summary": {
"action": "review",
"review_reason": [
{ "id": "pii_email", "score": 1, "threshold": 1 }
]
},
"ref": "comment_9034"
}What it checks
23 text categories in four groups
A workflow picks the categories you care about and the thresholds for each. The groups are only how this page lists them.
Harmful content
- Unsafe text
- Violence
- Sexual content
- Hate speech
- Illegal activity
- Unethical content
- Self-harm
- Child safety
- Profanity
- Extremism
Personal data
- Personal data
- Privacy violation
- Email address
- Phone number
- Bank account
- Payment card
Spam and scams
- Spam
- Scam
- Off-platform contact
Links
- Link
- Adult site link
- Gambling site link
- Unsafe site link
Rules on the text itself
Email addresses, phone numbers, bank account numbers, payment cards, links, spam and scams are found by rules and by lists of sites. They score 1 when they match, so their thresholds are 1.
Scores for the rest
Unsafe text, violence, sexual content, hate speech, illegal activity and similar categories get a score from 0 to 1, and you set a review and a reject threshold for each.
Comments, chat and AI output
The same call for people and for models
Check a comment before it is published, or the text an LLM generated before you send it. The decision and the reasons have the same shape, and the ref you send ties the decision to your comment or your agent run.
AI-generated content
The text an agent or an LLM generates, checked before you publish it, and the messages a user sends it.
The answer
Every decision explains itself
This text holds an email address, which the default workflow sends to review. The score is 1 because a rule found it.
{
"status": "success",
"workflow": "default",
"workflow_version": 1,
"summary": {
"action": "review",
"reject_prob": 0,
"reject_reason": [],
"review_reason": [
{
"id": "pii_email",
"text": "Email address",
"message": "This content contains an email address.",
"score": 1,
"threshold": 1,
"source": "rules"
}
]
},
"scores": { "pii_email": 1 },
"time_ms": 4,
"request_id": "0192b7c5-1d20-7b3a-8f61-5c9e2a7d4b10",
"ref": "comment_9034"
}Limits
What you can plan with
- Request
- JSON, with Content-Type: application/json
- Length
- Up to 20,000 characters, 5,000 on the free plan
- Your id
- A ref of 1 to 128 characters, echoed in the answer
- What counts
- One text is one check, whatever the number of categories
- What is kept
- The decision and your ref. Never the text.
In the docs
Read how it works
- Texts Check a text with POST /v1/text, for comments, messages, names and bios.
- Workflows Categories, thresholds, custom categories and versions.
- Responses and errors Every field of an answer and every error type.
- Limits and quotas What each plan includes, how fast you may call the API and what happens at a limit.
- Best practices Server-side only, fail closed, and messages chosen by category id.
- The journal and ref What the journal keeps, and how your own id finds a decision.
Questions
Before you start
- Which categories does a text check cover?
- Unsafe text, violence, sexual content, hate speech, illegal activity, unethical content, self-harm and child safety; personal data, from email addresses to payment cards; spam, scams and off-platform contact; and links to adult, gambling and unsafe sites. The list above is the whole set the shipped workflows use.
- Can it find personal data in a text?
- Yes. Email addresses, phone numbers, bank account numbers and payment card numbers are found by rules on the text itself, and a personal data category covers other cases.
- Can I check text an AI model wrote?
- Yes. The call is the same as for a comment: send the generated text before you publish it and act on the decision. A ref such as the run id ties it to your agent run.
- How long can a text be?
- At most 20,000 characters, and 5,000 on the free plan. A longer text is refused with 413 text_too_long: split it at paragraph boundaries and check each part, each part being one check.
- Do you store the texts?
- No. A text is checked in memory and forgotten when the answer is sent. The journal keeps the decision, the reasons and your ref.
Start with 2,000 free checks
No card needed. Create a key and send your first request in five minutes.
Beta prices, in euros, excluding VAT