Fillers, i.e. devices used to fill pauses in disfluencies, are underinvestigated throughout the world (Amiridze et al., 2010). Tupi-Guarani languages, a branch of the Tupi family spoken in Amazonia, are not an exception (Jensen, 1998; Rose, to appear). In contrast, there is a rich literature about the Tupi-Guarani word for ‘thing’, reconstructed as *maɁe (Jensen, 1998: 536). Two articles focus on the multiple grammaticalizations of this word beyond its basic referential nominal use (Dietrich, 1994; Auwera & Krasnoukhova, 2021). The latter sporadically mentions some discourse uses of *maɁe, such as a filler in Guarani translatable as ‘whatchamacallit’. Semantically bleached nouns with a generic meaning are in fact a known source for fillers (Podlesskaya, 2010). Building on this literature, the present talk intends to open an avenue for the research of *maɁe as a filler in Tupi-Guarani languages. In fact, Tupi-Guarani dictionaries, reference grammars, and texts give sparse and shallow information on the filler use of *maɁe (see for example Gasparini, 2015; Godoy, 2020; Grenand, 1989: 267). This paper offers a first detailed analysis of the discourse uses of *maɁe in a Tupi-Guarani language, namely Teko (emer 1243), in the hope that it could serve in the future as a basis for comparison within the family. The analysis of the Teko baʔe reflex of *maʔe is based on a corpus of spontaneous texts (>5000 words) collected by the author in the field. This word form is mainly used as a noun for a non-human entity (a thing, an animal or a supernatural being - note that the usual translation as ‘thing’ is not exact). It is also used as a verb ‘to do, to make’. The same form baʔe also endorses other grammaticalized functions highlighted in the literature, as a generic noun, a compound formative, and an interrogative pronoun. As for discourse uses, baʔe is essentially used as a filler in the texts, with 65 such tokens being found (12 / 1000 words). When used as a filler, it is found without any morphology, an evidence for reanalysis. It most often delays the production of a noun (including some with human referents. It is less often found before verbs or other parts-of-speech. Prosodically, several aspects of its realization as a filler differ from those as a noun: the average durations of its final vowel, the preceding pause, and the next word are statistically significantly higher (data include 23 tokens as fillers and 23 as nouns). Three other discourse uses, not participating to disfluency, had never been mentionned before in the literature : with a plural suffix as a general extender after a list meaning ‘etc’, as a first (euphemistic) approximation before a reference to a supernatural being, and with an injunction clitic, as an attention-getter, meaning ‘sorry ! watch out !’. The theoretical goal of this paper is to complement the map of grammaticalization paths of Tupi-Guarani *maʔe sketched by van der Auwera and Krasnoukhova (2021: 90) by suggesting additional paths of pragmaticalization (Diewald, 2011). For example, we hypothesize a path of grammaticalization from baʔe as a generic noun to its filler use, which could in turn have later lexicalized into the verb ‘do, make’ (found only in Teko). This paper will therefore put forward hypotheses for a better understanding of the synchronic use and historical development of fillers, that could further be tested on other Tupi-Guarani languages.