Recently, the local friendliness (LF) no-go theorem has attracted considerable interest due to its deep foundational implications. This no-go theorem applies to scenarios which combine Bell experiments with Wigner's friend-type setups, containing spacelike separated superobservers assumed to be capable of performing quantum operations on a local observer, also known as their “friend.” The theorem shows that a set of assumptions termed “local friendliness”—weaker than the set of assumptions required for proving Bell's theorem—constrains the space of probabilistic behaviors in extended Wigner's friend scenarios to the “local friendliness (LF) polytope,” and that quantum correlations can, in principle, violate those constraints. In this work, we extend the study of local friendliness to multipartite scenarios—with arbitrary but finite numbers of superobservers, friends, measurements, and outcomes—and fully characterize the corresponding LF polytopes, their relationships with each other, with the Bell polytope, the no-signaling polytope, and the quantum set of correlations. In particular, we identify, via constructive proofs, all the canonical LF scenarios where the LF polytope coincides with the Bell polytope—hence identifying all the scenarios where “genuine LF inequalities” may arise. Among other results, we also show that whenever the LF polytope is distinct from the Bell and the no-signaling polytope, it also is neither contained in nor contains the quantum set of correlations. Apart from the foundational motivation of better understanding the constraints from local friendliness, our results may also find use in a priori unrelated contexts of device-independent information processing, where the LF polytopes have also been shown to arise, under the name of “partially deterministic polytopes.”