Abstract This article discusses object (and subject) omission in three Guaraní languages, providing evidence for agreement–based direct object drop and genuine pro in internal argument position. The empirical facts show that 1p and 2p object omission behaves very similarly to classical, even consistent, and agreement–based subject drop in better-studied languages (Barbosa, 2011a; Roberts & Holmberg, 2010). On the other hand, we see that Guaraní languages additionally allow for discourse-anchored 3p null arguments, occurring in parallel fashion in both external and internal positions. These patterns approximate Guaraní-style null argument licensing to what was previously described for radical pro-drop languages (Barbosa, 2011b; Holmberg, 2005; Huang, 1984; Tomioka, 2003). Hence, the resulting system referred to here as person-split pro-drop integrates both agreement–based pro-drop and the licensing of agreement-independent argument omission, in a system where several argument omission types not only co-exist but also interact in mixed argument configurations. Core to the pattern is the asymmetric null argument licensing mainly along the lines of 1/2p(erson) versus 3p features rather than along the line of grammatical function. Even though I differentiate two classes of empty nominals in the system, namely weak pronouns and empty nominals [NP e], this distinction alone cannot fully account for the Guaraní-style pro-drop pattern. Central to modelling argument omission encountered in the Guaraní data is to employ theoretical proposals on the role of highly articulated agree probes [u3, u2, u1] — generating sensitivity to Person Hierarchy (PH) effects in the (narrow) syntax — into the overall generative debate on null argument licensing (Béjar, 2003; Béjar & Rezac, 2008; Oxford, 2014, 2017, 2019).