Moderation API · VigiMod 1.2
Moderation that
reads the room.
VigiMod flags toxic, hateful, sexual and spammy messages in 26 languages. It reads the conversation around each message and checks the rules your community writes in plain words.
- built-in categories
- 8
- built-in categories
- training languages
- 26
- training languages
- per message on a 2-core CPU
- ~1 s
- per message on a 2-core CPU
- messages stored
- 0
- messages stored
Live demo
Try it on your own words
The real model, the real scores. Nothing you type is stored.
Scores appear here.
Scores go from 0 to 1. A category is flagged when its score passes its threshold, and every server can set its own thresholds.
In a real server
The bot decides. VigiMod tells it what it saw.
Real answers of VigiMod 1.2 to these messages. What happens next is up to you: delete, warn, or send to a moderator.
Server rules sent with every message
- No spoilers about movies or series
- No buying, selling or trading items
mira
anyone watched the finale last night?
tobi
not yet, tonight!
kaz
lmao the dog dies at the end
no-spoilers 1.00· after the chat aboveluca
seguite il mio canale, link in bio!!
Advertising 1.00yuki
このゲーム最高！また明日やろう
✓ okjonas
Halt die Klappe, du Idiot
Toxic 1.00sam
gg wp, that last round was insane
✓ okana
who wants to trade skins? I have the dragon one
no-trading 0.97
Why VigiMod
Built for communities, not keyword lists
- 01
Reads the conversation
Send up to 10 previous messages with each one. The same words can be harmless or a spoiler, a joke or a threat, depending on what came before.
- 02
Your rules, in plain words
Write rules like “No trading items” or “No spoilers”. Every message is checked against them, up to 10 rules per request, in any language.
- 03
Settings per server
Pick the categories, the sensitivity of each one and of each rule, and the kinds of slurs to block, with every request. Nothing to set up on our side.
- 04
Slurs, by kind
Block racial, xenophobic, antisemitic, homophobic, transphobic or ableist slurs separately, so each server can treat them differently.
- 05
Spam and floods
Send the user id and VigiMod spots repeated messages and floods per user and per channel.
- 06
We don't keep your messages
Messages are scored and forgotten. We only count requests and flags per hour, so your dashboard can show your usage.
One request
Send a message, get scores and flags
One HTTP call with your key. Add the previous messages, your rules and your thresholds when you need them.
curl -X POST https://devcalledalex-vigimod-api.hf.space/moderate \
-H "X-API-Key: $VIGIMOD_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "lmao the dog dies at the end",
"context": ["anyone watched the finale?", "not yet, tonight!"],
"rules": {"no-spoilers": "No spoilers about movies or series"}
}'Response
{
"flagged": true,
"flags": { "rule:no-spoilers": true, "toxic": false, "threat": false, … },
"scores": { "rule:no-spoilers": 0.998, "toxic": 0.006, "threat": 0.003, … },
"model_version": "vigimod-1.2"
}Languages
Trained in 26 languages
Most of the training data is in Italian, English, Spanish, French, German and Portuguese. The model underneath was pre-trained on about 100 languages.
Italiano · English · Español · Français · Deutsch · Português · Nederlands · Polski · Română · Türkçe · Русский · Українська · Čeština · Svenska · Ελληνικά · Magyar · العربية · עברית · فارسی · हिन्दी · 中文 · 日本語 · 한국어 · Tiếng Việt · Bahasa Indonesia · ไทย
Accounts open soon
Sign in with Discord, GitHub, Google or email, create your keys and follow your usage. The API is already running.
Try the demo