openshield
New Fraud prevention, in the same account

Content moderation and fraud prevention APIs

Moderate what is posted. Stop the fraud behind it.

One API call checks an image or a text and answers accept, review or reject. Another scores a sign-up, a login or a payment and answers allow, review or block. One account, one balance, rules that tune themselves.

Get started freeRead the docs

Up to 2,000 free checks every month. No card needed.

openshieldModerationJournal

Every post, checked before it goes live

Images and texts get accept, review or reject, with the category and score behind each one.

TimeContentDecisionReasons
14:02:11avatar_4821Imagerejectnudity 0.97
14:02:09msg_9034Textreviewpii_email 1
14:02:06comment_7716Textaccept-
14:02:03upload_3390Imagereviewweapon 0.62
14:01:58bio_2207Textaccept-
Request
curl -X POST \
  -H "Authorization: Bearer $OPENSHIELD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Write me at jane@example.com",
    "ref": "msg_9034"
  }' \
  "https://api.openshield.dev/v1/text"
Answer, trimmedreview
{
  "summary": {
    "action": "review",
    "review_reason": [
      { "id": "pii_email", "score": 1, "threshold": 1 }
    ]
  },
  "ref": "msg_9034"
}
  • 2,000free checks every month at most, with the full answer
  • 0images or texts kept after a check
  • EUserver and databases in the European Union

Built for the agents era

Your users post. Your agents write.

Check both with the same call, the same rules and the same journal. What differs is where the content comes from.

For your SaaS

Moderate what your users post

Posts, comments, avatars and uploads, checked before anything goes live.

  • SightEngine-compatible workflow routes: change a base URL and a key
  • Workflows per part of your product, with keys limited to the ones they need
  • A journal you search by your own id, with no content in it

See how it works for your SaaS

For your AI agents

Moderate what your agents read and write

The messages and images users send an agent, and the text and images it generates, checked before they are published.

  • One ref ties each decision to the agent run or message
  • A workflow per agent, stricter or looser, and a key limited to it
  • Agent-ready: an integration prompt, llms-full.txt and the OpenAPI document

See how it works for your agents

Fraud prevention

Stop the fraud behind the content

Score every sign-up, login and payment before you act. IP, email, phone, device and velocity signals, one call, and little friction for real users.

One call

Detect, score, and act

Ask before a sensitive action. Get allow, review or block, with the reasons that fired.

Automatic

Rules that turn themselves on

Rules start by watching. After 30 days, the ones proven on your labels act on their own.

Safe by default

A brake and a way back

Stop a sudden wave of blocks from locking out users. Prefer to decide yourself? Switch to manual.

Your own instance

Pseudonymized, in Germany

Get an openfraud instance for your organization. Events are pseudonymized and kept 30 days, 90 days from your first top-up.

Fraud prevention API Fraud prevention

How it works

Live in five minutes

Nothing to set up after sign-up: your account starts with a ready workflow, and the dashboard walks you through the rest.

  1. 01

    Sign up

    Email and password, a magic link, GitHub or Google. Your account comes with a working default workflow.

  2. 02

    Create a key

    One click in the dashboard. The key is shown once: keep it in OPENSHIELD_API_KEY on your server.

  3. 03

    Paste the prompt

    Give the integration prompt to your coding agent. It reads our docs and the OpenAPI document, and adds the check to your code.

Read the quickstart

What you get

One call, every reason

Everyone, before a first top-up too, gets the same full answer: the decision, the categories behind it and their scores.

One request

A decision you can act on

Send the bytes or a text. Get accept, review or reject, with the categories that decided it.

Categories

Built-in categories, and your own

Images and text are checked against the categories you keep in a workflow. Describe your own image categories in plain words.

Images

  • Nudity
  • Suggestive content
  • Gore
  • Violence
  • Weapon
  • Drugs
  • Hate symbol
  • Self-harm
  • and your own

Texts

  • Unsafe text
  • Hate speech
  • Sexual content
  • Violence
  • Illegal activity
  • Personal data
  • Spam
  • Scam

All image categoriesAll text categories

Nothing stored

The check leaves no copy

Content is checked in memory and forgotten. The journal keeps the decision, never the content.

Workflows

Your rules, versioned

Set a review and a reject threshold per category. Every change is a new version you can compare.

SightEngine compatible

Switch with a new base URL

Three SightEngine workflow routes with the same paths, request fields and answer envelope. Category ids and model fields differ: the docs map them.

SightEngine alternative

Open source

Built on openmod, the open-source engine

The engine is public under the Apache-2.0 licence, and you can run it yourself. openshield is the hosted API: you run nothing.

openmod on GitHubGitHub, opens in a new tab

Pricing

Pay for what you check

No subscription. Every check takes what its workflow costs from a prepaid balance, and every account gets free use each month.

Images

Per image, by what its workflow checks

Nudity and sexual content
€0.001
Visual categories
+ €0.001
Text in images
+ €0.002
QR codes
+ €0.001

A usual workflow costs €0.002 an image; everything on, €0.005.

Texts

Per started 1,000 characters

Rules and links
€0.001
AI models
+ €0.002
Advanced reading
+ €0.002

A short text costs €0.001 with rules only, €0.003 with the AI models.

Fraud prevention

Per call, with your own openfraud instance (beta)

Decision
€0.001
Event
€0.0002
IP lookup
€0.0002
Network data, when on
+ €0.0002
Activity bursts, when on
+ €0.0005

10,000 sign-ups scored cost €10.00. Cases, accounts and lists cost nothing.

See the prices in detail

Questions

Before you start

Can I moderate what my AI agents write?
Yes. Send the text or the image an agent generated before you publish it, and the message or the image a user sends the agent before it reaches the model. Each call answers accept, review or reject with the reasons, and a ref ties the decision to the agent run.
Do you store my images or texts?
No. Content is checked in memory and forgotten once the answer is sent. The decision journal keeps metadata only: the time, the decision, the reasons with their scores, and the reference you sent. The fraud prevention API is the one exception, since a fraud decision needs a history: its events, decisions and cases are kept 30 days until your first top-up and 90 days after, with email addresses and IP addresses replaced by pseudonyms.
What does a check cost?
You pay for what you check, from a prepaid balance with no subscription. A usual image workflow costs €0.002 an image, and a text checked by the AI models €0.003 per started 1,000 characters, with every category included. Every account gets up to 2,000 free checks each month (€2.00 of use).
Can I decide what is acceptable?
Yes. Each workflow sets a review threshold and a reject threshold per category, and you can describe image categories of your own in plain words. Every change makes a new version.
Can I migrate from SightEngine?
If you use SightEngine's workflow routes, yes. The image check (POST and GET) and the text check keep the same paths and request fields, and the answer carries the same envelope and summary. Point your base URL at us, send an openshield key as api_secret and map the category ids. SightEngine's other endpoints and its per-model fields are not served.
Is it an open-source content moderation API?
The engines are open source: openmod, for moderation, and openfraud, for fraud prevention, are under the Apache-2.0 licence and you can run them yourself. openshield is the hosted service built on them, run by Sway SRL in Belgium. We do not sell a self-hosted version.
Where does it run, and what about GDPR?
The moderation server and the databases are in the European Union, and the fraud prevention instances run on a server in Germany. Cloudflare sits in front of the site and the API to deliver and protect them. For the personal data your content may hold, Sway SRL is your processor, and moderated content is never stored. The privacy policy, the data processing agreement and its list of sub-processors are on this site.
What does openshield not do?
The moderation API checks images and text. It does not check video or audio, it does not detect prompt injection, and nobody reviews content for you: a review decision goes back to your own team.
Does openshield also prevent fraud?
Yes, as a second product. The fraud prevention API scores a sign-up, a login or a payment before you act and answers allow, step up, hold or block, with the rules that fired. It runs openfraud, the open-source engine, with one instance for each organization, from €0.001 a decision.
Can I call the API from a browser?
No. Keys are secret and belong on your servers. The API refuses calls that carry a browser Origin header, and the docs show how to call it from your backend.

Start with up to 2,000 free checks every month

No card needed. Create a key and send your first request in five minutes.

Get started free

Prepaid, in euros, excluding VAT