On 10/28/2016 3:15 AM, Nicola Piazzi wrote:

I use the same box for internal mail delivery

I shortcircuit internal messages that come from internal ip

I noticed that bayes never learn from internal messages if I take one and make sa-learn –ham of these messages it answer that have learned

Is possible to learn automatically ?

The first thing to do is see why they are not being learned.  Look at the X-Spam-Status header and make note of the autolearn status.  You can get more info as well as a list of the possible states for autolearn here: