For marketplaces
Moderate what sellers publish
Listing photos, descriptions and new seller accounts: one API call before anything goes live. You get accept, review or reject on content, or allow, step up, hold or block on a sign-up, with the reasons.
2,000 free credits every month, up to 1,000 images. No card needed.
curl -X POST \
-H "Authorization: Bearer $OPENSHIELD_API_KEY" \
--data-binary @photo.jpg \
"https://api.openshield.dev/v1/image?ref=avatar_4821"{
"summary": {
"action": "reject",
"reject_reason": [
{ "id": "nudity", "score": 0.97, "threshold": 0.8 }
]
},
"scores": { "nudity": 0.97, "gore": 0.01 },
"ref": "avatar_4821"
}What you check
Everything a seller can publish
One API for listings and messages, and one for new seller accounts, from the same account.
Listing photos and attachments
JPEG, PNG, WebP or GIF, up to 15 MB. Nudity, weapons, drugs, gore and hate symbols are built-in categories. Text written on an image and the links in QR codes can be read and checked, including scam and unsafe-site links. You can add your own categories in words, such as a prohibited item in your catalogue, without training a model: up to 20 custom categories per workflow.
Titles, descriptions and seller messages
Titles, descriptions, reviews and thread messages go through the text API: spam, scams, off-platform contact, personal data, and links to adult, gambling or unsafe sites. The same call works for a listing and for a message to a buyer.
New seller accounts
Fake or farmed seller accounts are not a content problem. The fraud prevention API (beta) scores a sign-up, a login or a payment: allow, step up, hold or block, with the rules that fired. Pack rules start in shadow. Same account and prepaid balance as moderation.
Why it fits a marketplace
A decision your code can act on
Accept, review or reject
Accept: publish the listing. Review: hold it for your team. Reject: refuse it, with a message you choose from the category id. openshield does not review listings for you.
A workflow per surface
Photos, descriptions and chat each get a workflow, with its own categories and review and reject thresholds. Every change is a new version you can compare.
A key per service
Limit a key to the workflows that surface needs. Rotate it with a grace delay, so a deploy never has a gap.
A journal you can search
Each decision is kept with your listing or seller id (ref), the reasons, the scores and the workflow version: 7 days before a first top-up, 30 days after. The content is never kept.
Nothing stored, in the EU
Content is checked in memory and forgotten when the answer is sent, in the European Union. Fraud instances run in Germany, and their events are pseudonymized.
Seller sign-up abuse too
Fake seller accounts and sign-up farms are not content. The fraud prevention API scores a sign-up before you create the account, from the same account and balance.
How it fits
Check before a listing goes live
Your server calls the API with your listing id, then publishes, holds or refuses. For a new seller, call POST /fraud/v1/decide before you create the account (beta).
- 01
A seller publishes
A seller uploads a photo or saves a description. Your server gets it, as it does today.
- 02
Check it
Call POST /v1/image or POST /v1/text with the content and your own ref, the listing id. The answer carries the decision and the reasons.
- 03
Publish, hold or refuse
Act on summary.action. If the check fails or times out, hold the listing: do not publish what was never checked.
import { checkText, ModerationError } from "./moderation"
// A request the API refuses as it is: it will never be accepted.
const refused = [400, 413, 415]
export async function moderateComment(text: string, commentId: string) {
try {
const decision = await checkText(text, commentId)
if (decision.action === "accept") return "publish"
if (decision.action === "review") return "hold"
return "refuse"
} catch (error) {
const status = error instanceof ModerationError ? error.status : 0
if (refused.includes(status)) return "refuse"
// A timeout or any other failure: keep it hidden for now.
return "hold"
}
}Categories
What a check can find in a listing
These come from openmod's catalogue. A workflow picks the ones you care about and the thresholds for each.
Images
- Nudity
- Suggestive content
- Gore
- Violence
- Weapon
- Drugs
- Hate symbol
- Self-harm
- QR code
- and your own
Texts
- Spam
- Scam
- Off-platform contact
- Email address
- Phone number
- Adult site link
- Gambling site link
- Unsafe site link
- Hate speech
- Sexual content
- Illegal activity
In the docs
Read how it works
- Quickstart Your first check in five minutes, from an account to a key and one request.
- Best practices Server-side only, fail closed, and messages chosen by category id.
- Workflows Categories, thresholds, custom categories and versions.
- The journal and ref What the journal keeps, and how your own id finds a decision.
- Node.js and TypeScript A small module with timeouts and a decision that fails closed.
- Fraud prevention Decide and events, the end user's context, prices, webhooks and errors.
- Migrating from the OpenAI moderation API Call POST /v1/text or POST /v1/image, then map the decision and the category ids.
More
Other ways in
- For your SaaS Moderate posts, comments, avatars and uploads before they go live.
- For your AI agents Moderate what an agent reads and what it writes.
- For dating apps Moderate profile photos, bios and messages before anyone else sees them.
- Image moderation API Nudity, gore, weapons and more, with a score per category.
- Text moderation API Unsafe text, personal data, spam and scams, with the reasons.
- Fraud prevention API Sign-up farms, card testing and account takeover, with your own openfraud instance.
- SightEngine alternative The same workflow routes: a new base URL and a new key.
- Pricing Free use each month, then a prepaid balance, priced by what each check uses.
Questions
Before you start
- What do I do with the review outcome?
- Hide the listing or the message from other users until someone on your team approves it. If you have no moderation queue, start with a flag on the row and a screen that lists the flagged rows. openshield does not review content for you.
- What if a check fails or times out?
- Hold the listing until a retry succeeds or someone has looked at it. Publishing what was never checked is the case a moderation API exists to prevent.
- Can I call the API from my front end?
- No. Keys are secret and belong on your servers, and the API refuses calls that carry a browser Origin header. Call it from your backend and give your front end the decision.
- Do you store listing photos or descriptions?
- No. Content is checked in memory and forgotten once the answer is sent. The journal keeps metadata only: the time, the decision, the reasons, the scores and your ref. Fraud events are the documented exception, kept pseudonymized: see Privacy and data in the docs.
- What does a check cost?
- Beta prices, in euros, excluding VAT, from a prepaid balance with no subscription: €0.002 for an image under a usual workflow (nudity and visual categories), €0.003 per started 1,000 characters of text checked with rules and AI, and €0.001 for a fraud decision. Every account gets 2,000 free credits each month (up to 1,000 images), shared by moderation checks and fraud calls. It does not carry over to the next month, and no card is needed. The full grid is on the pricing page.
Start with 2,000 free credits every month
No card needed. Create a key and send your first request in five minutes.
Prepaid, in euros, excluding VAT