ModerBot reads the whole message: the topic, the tone and what was said before it. That is why a question to the group stays a question, while the same word in a sales pitch is an ad.
Four steps, and far from everything reaches the model: it is needed where meaning decides the outcome, not where a string matches.
Look-alike letters from other alphabets, dots inside words, invisible characters and Unicode fonts are unfolded back into plain text.
Word lists, links, attachment types, the author's behaviour. Most messages are settled here, instantly.
Where intent decides the outcome, the model steps in: it reads the whole phrase and takes the conversation into account.
The action chosen for that category is applied, and the decision with its reason goes into the log.
A spammer posts, the bot takes the message down and explains what for. The notice can be shortened or turned off entirely.
The conversation is a reconstruction: the names and the wording are made up, the format of the notice is the real one.
The same word, two different intentions. To a list of banned words these messages are indistinguishable.
Eight groups, backed by 37 categories — each one switched on separately.
Repeated messages, bulk mailings, an avalanche of messages from one person.
Commercial offers, invitations to other chats and channels.
No direct offer, but a clear intent to move the conversation into DMs.
Fraud schemes, fake links, attempts to fish out personal data.
Personal attacks, aggression, obscene language — if the category is on.
Drugs, adult services, extremism, violence, doxxing.
Text is read off the picture and checked like an ordinary message.
Broken-up words, digits for letters, invisible characters, odd fonts.
A different one for every category. One violation is worth a warning, another an immediate ban.
Automatic moderation gets things wrong — all of it does. The difference is whether you can see it and undo it.
Unban, lift a restriction or add an exception — straight from the row in the log.
A member or a domain the rules do not touch at all. Added from the panel or with the /whitelist command.
Owners, admins, posts sent on behalf of the linked channel and anonymous admins are never checked.
ModerBot is free for groups of any size. Critical categories switch on right away; the rest you tune to your chat.
Add to Telegram