Abstract: Animals constantly sample information from the external world, and memorizing some of it is essential for adaptive behavior. Classical views suggest that memories are stored in assemblies of excitatory neurons and retrieved through attractor dynamics. However, increasing evidence indicates that during learning, groups of both excitatory (E) and inhibitory (I) neurons become strongly interconnected, forming E-I assemblies. To investigate the effects of these assemblies on memory function, we built a biologically constrained spiking neural network model of a brain region involved in olfactory memory. We first studied how introducing E–I assemblies reorganizes odor-evoked activity patterns in neural state space. We found that, unlike classical E assemblies, E-I assemblies do not generate attractor states. Instead, they induce changes in neural activity that promote both stability and flexibility, enhancing the retrieval of learned odors while preserving the ability to discriminate novel odors. By partially manipulating inhibitory neurons, we further showed that inhibitory neurons within assemblies prevent networks from responding similarly to dissimilar odors. These findings provide valuable insights into how specific E-I circuit connectivity affects the storage and retrieval of information. Furthermore, model predictions were confirmed experimentally, supporting the existence of E-I assemblies in biological networks. We are currently expanding this work to multi-region models to understand how memories are stored across distributed brain regions and how these regions communicate and coordinate during memory retrieval.