Keep spam out of your Telegram group without captchas
Add @jev_antispam_bot as an administrator and it checks every message from non-admins with Jev, a spam model made for chat messages. It deletes only what it's confident about, like crypto shilling, phishing, fake support and "check my bio" funnels, and leaves ordinary conversation alone.
Open source (MIT) and free. There are no captchas and nothing for members to do.
How to use it
- Add @jev_antispam_bot to your group.
- Make it an administrator with the Delete messages permission.
- Send
/statusin the group. It answers "Active" when it can delete spam there. - That's all. New and edited messages from members are checked from then on.
What it deletes and what it leaves
| Message | Verdict |
|---|---|
| Token or casino shilling | Deleted |
| Phishing, fake airdrops | Deleted |
| "Check my bio" or DM funnels | Deleted |
| Mass mentions, repeats | Deleted |
| Messages from admins | Never checked |
| Ordinary conversation | Kept |
| When unsure | Kept |
A spam pitch split across several short messages is caught as one, and an account that sends the same pitch to several groups gets its earlier copies removed too. If the model or Telegram can't be reached, messages stay where they are.
Questions
What does it need from me?
Administrator rights with the Delete messages permission. Nothing else: there are no settings to tune and no captcha for new members.
Will it delete normal messages?
It deletes a message only when Jev's spam score is at least 0.81, and it skips administrators and your group's own linked channel. No filter is perfect, so a borderline message can occasionally go, but the bot is tuned to keep anything it's unsure about.
Does it ban or mute people?
No. It only deletes spam messages. Banning stays with you.
What happens to the messages it reads?
To get a verdict, it sends the message text (and, for short messages, the sender's public bio) to the Jev API by TypeSafe. The bot doesn't store or log message text; it keeps the last few messages per sender in memory for about ten minutes to spot split pitches.
Which languages does it understand?
The model reads spam in many languages; the bot is tested on English, Russian and Chinese spam. The bot's own replies are in English.
Is it free?
Yes. There is no paid plan, and the code is open source under the MIT license.
Languages
- English (replies)
- Spam in many languages, tested on English, Russian and Chinese