AI moderation

Rules catch what you described in advance. The trouble is that spammers reword things faster than you can edit your rule sets: yesterday "work from home", today "part-time in your spare time", tomorrow the same thing as a picture. This release adds a block that looks at what a message means rather than at its letters.

Six topics instead of word lists#

The AI moderation block hands a message to a language model, and setting it up comes down to checkboxes: you tick the topics that count as violations in your chat.

Topic	What it covers
Illegal substances	selling and brokering, dead drops and everything around them
Gambling ads	casinos, betting, slots, bonus funnels
Crypto scams	pumps, fake airdrops, "guaranteed profit", P2P fraud
Phishing	fake logins, impersonated admins and support, account takeovers
Job scams	easy money, prepayment schemes, money-mule "jobs"
Adult spam	selling sexual services, escort ads, erotic promos

No word lists, no patterns: the model recognises a topic in any wording, including one nobody thought of in advance.

One detail matters: a punishment is applied only on high confidence. The model still labels borderline messages, but they simply go into the log — you can see it noticed something, and nobody gets punished for it.

Photos#

A photo is analysed together with its caption, so an ad on a banner and a QR code on a screenshot stop being a blind spot. Stickers, videos, voice notes and video messages are not processed. You can check text only, photos only, or both — whatever suits the chat and the budget.

Balance and history#

This is the only block that spends money as it works: you top up a balance in advance and it is drawn down for the messages actually checked. The block does not depend on the chat's plan — it runs as long as the balance is positive.

The history tab shows the latest transactions, the average cost per message, how much went out over thirty days, and an estimate of how many more messages and days the remainder covers. New chats get their first dollar from us: try the model on your own traffic rather than on our promises.