rspamd and dovecot antispam

jca

Verified User
Joined
Oct 31, 2006
Messages
306
Location
Allen, TX
Has anybody integrated this guide with DA?

Part1: https://kaworu.ch/blog/2014/03/25/dovecot-antispam-with-rspamd/
Part2: https://kaworu.ch/blog/2015/10/12/dovecot-antispam-with-rspamd-part2/

There also seems this guide that looks a little more straightforward: https://rspamd.com/doc/tutorials/feedback_from_users_with_IMAPSieve.html

Seems like a great idea to integrate learning to rspamd and junk folders from customers, so they could actively help the algorithm learn.

Guides look a little bit old, but were taken directly from the rspamd documentation: https://rspamd.com/doc/tutorials so not sure how they might differ now.

Any tips from anybody doing this or anything similar already in DA? Personally I think it would be huge to be integrated to DA/CustomBuild for more active Spam control.
 
Nice, I like the roundcube idea, but I think it would be a better/more complete approach to do it via Sieve/IMAP. That way if user is using IMAP client it would also support reporting of SPAM vs only via Roundcube. Still any real/direct way for users to report spam would be ideal to strengthen the anti spam protection.
 
Does anyone have a guide for implementing teaching rspamd using IMAP/Roundcube?

I can find guides such as https://doc.dovecot.org/configuration_manual/howto/antispam_with_sieve/ but I don't know if anything needs to be changed for it to work with DirectAdmin?
I already have autolearning working with rspamd as it seems to do its job.

Basically I would like to teach rspamd as soon as a user moves mails to the Junk folder, either using their prefered IMAP client or using Roundcube. And if a mail is moved out of the Junk folder and to any other folder, then it would be taught as ham.
 
Autolearning from mail moved into the junk folder would be relatively easy I suppose. The other way around might be a bit resource heavy if done on large installs.
 
In theory it would only happen when you move spam out of the junk email out to anywhere (except trash) so it should not be too resource heavy. Still would be a great way to allow customers to train the own algorithm.
 
Back
Top