Autobigraphical memory, our memory for events, may employ a code that economically uses existing knowledge. This helps us remove redundancies: we can store a bit less, because we can use background knowledge to reconstruct parts of the events. In this theoretical paper (Mazor et al, Neurosci Biobeh Rev 2026), we propose that our memory makes efficient use of a “self model”: knowing what we would do in given situations. This means we can reconstruct events, by reconstructing what we would have done. We make testable predictions about healthy behaviour and fMRI, Alzheimer’s disease, Korsakoff’s dementia, and obsessive compulsive disorder.