Abstract: In the last decades North African desert ants of the genus Cataglyphis Foerster, 1850 – and more recently their ecological equivalents in the Namib desert (Ocymyrmex Emery, 1886) and Australia (Melophorus Lubbock, 1883) – have become model organisms for the study of insect navigation. While foraging individually over distances of many thousand times their body lengths in featureless as well as cluttered terrain, they navigate predominantly by visual means using vector navigation (path integration) and landmark-guidance mechanisms as well as systematic-search and targetexpansion strategies as their main navigational tools. In vector navigation they employ several ways of acquiring information about directions steered (compass information) and distances covered (odometer information). In landmark guidance they rely on view-based information about visual scenes obtained at certain vantage points and combined with certain steering (motor) commands of what to do next. By exploring how these various navigational routines interact, the current position paper provides a hypothesis of what the architecture of the ant's navigational toolkit might look like. The hypothesis is built on the assumption that the toolkit consists of a number of domain-specific routines. Even though these routines are quite rigidly preordained (and get modified during the ant's lifetime by strictly task-dependent, rapid learning processes), they interact quite flexibly in various, largely context-dependent ways. However, they are not suited to provide the ant with cartographic information about the locations of places within the animal's foraging environment. The navigational toolkit does not seem to contain a central integration state in which local landmark memories are embedded in a global system of metric coordinates.