In the study of distributed systems, the assumption —commitment framework is crucial for compositional specification of processes. The idea is that we reason about each process separately, making suitable assumptions about other processes in the system. Symmetrically, each process commits to certain actions which the other processes can rely on. We study such a framework from an automata-theoretic viewpoint. We present systems of finite state automata which make assumptions about the behaviour of other automata and make commitments about their own behaviour. We characterize the languages accepted by these systems to be the regular trace languages (of Mazurkiewicz) over an associated independence alphabet, and present a syntactic characterization of these languages using top-level parallelism. The results smoothly generalize for automata over infinite words as well.