Main audio (microphone) goes into a Mic input.
Secondary audio (remote audio from a Tandberg audio codec, for example) goes into another input, and you turn the AUX knob for that channel to 0.
Then you take an AUX send back to your remote audio return input, so the caller does not hear his own audio, but hears everything else on the main mix.
Meanwhile the main mix contains all inputs, regardless of AUX settings, which goes to your record deck or PA or both.
Have I missed anything? Obviously I am interested in the theory in this post, not going to get into specific gear being used.