Small decisions. Interesting possibilities.Submit contentSubmit

Jev Anti-Spam Bot

A grammY Telegram bot that sends each group message with recent context to Jev as a set of yes/no spam questions and deletes only high-confidence matches.

Added to Jevfast

How it uses Jev

The state holds the group message and the recent messages around it for context. The questions are noul probes for separate spam classes, including unsolicited promotion, promotion amplified through a quote, profile bait, adult profile bait and recruitment pitches, and code deletes a message only when an answer crosses its configured threshold.

What Jev decides

One group-message check (illustrative)Example answers · not a recorded Jev response · Source ↗
Question 1 · unsolicited promotion
YOUR APP
INSTRUCTION

Is the message unsolicited promotional spam that a reasonable Telegram group moderator should delete?

STATE

The state is the message plus recent group messages for context. Each spam class is its own yes or no question in the same request, and the two shown are part of that set.

JEV · NOUL
YesNo
Question 2 · profile bait
YOUR APP
INSTRUCTION

Is the message spam that funnels readers to the sender's bio, profile, private messages, channel or a hidden link?

STATE

The state is the message plus recent group messages for context. Each spam class is its own yes or no question in the same request, and the two shown are part of that set.

JEV · NOUL
YesNo

App workflow

  1. Receive the message

    The grammY bot sees every group message once privacy mode is disabled and checks administrators, bot senders and verified linked channels for an exemption first.

  2. Build the state

    The message and recent messages become the state, with media button labels standing in for empty text when needed.

  3. Ask the spam questions

    One request asks each spam class as a separate yes or no question, with a cached or skipped call where policy allows.

  4. Delete or leave

    Only a high-confidence answer triggers a deletion, and a Jev or Telegram failure leaves the message untouched.

Why it is interesting

A moderator decision becomes a set of independent probabilities instead of one generated verdict, so a single confident signal can act while an uncertain one leaves the message alone. The exemptions for administrators, bots and a verified linked channel are settled in code before a request is spent, and a provider failure leaves the message untouched.