SpamFilter will analyze all emails that go thru it. SpamFilter is designed to handle incoming emails only. If you use SpamFilter to process outbound emails as well, you'll need to add an IP whitelist so that clients within your network will be allowed to use SpamFilter as a relay to send emails to the internet.

If this is done, SpamFilter will process the outgoing emails just like all the incomings, except that as the outgoing are all whitelisted, they will be marked as "good", and the statistical filter will "learn" that they are good emails.

However please note that all Bayesian filters work by comparing an incoming email against all the other incoming emails that have been received. All incoming emails you receive create some sort of "pattern" (we call it DNA sampling) that is unique to your installation. If you "interfere" with the flow of incoming emails by causing the bayesian filter to process your outgoing emails as well, this may altern the "pattern" and cause it to be different than it would be if only receiving emails.

As far as the quarantine is concerned, almost all spam will be saved in the database. The only exception are emails *not* addressed to your domains. These are always attempts by spammers to use your server as an open relay to bounce emails to other servers. As the emails are not addressed to your users, they are neither received nor saved.

The email body is stored in the Msg field in the tblMsg table in the database.

That cannot be done currently. In order to learn that an email that was delivered is acutally spam, a user would need to send the *original*, unmodified email to Spam Filter so it can be re-processed. However many email clients will greatly modify the email's content upon reading and forwarding. Inorder to achieve accurate results, the Bayesian filter will need to see the email exactly as it was received. Feeding into the filter an altered email will not produce accurate results. Due to the difficulty in ensuring that users forward the exact email contents, this feature has not been implemented.

Yes, that is the same option we've been looking at.

However it is our experience that the Bayesian filter, while it does its job, its not as effective as some of the other filters we use. Due to this, to tell the truth, we've been concentrating more on those, rather than adding that additional feature to the bayesian filter.