Since the invention of the Semantic Web concept, the number of ontological and terminological resources on the Web has been increasing. These heterogeneous resources may share entities or contain similar ones. To cope with the diversity of these resources and collect them, resources repositories have been created. Matching entities belonging to different resources is then required to use or combine them, which generates alignment resources. Thus, collecting and managing resulting alignments is useful in order to compose them, compare sets of alignments for the same input resources, enrich existing alignments from each other or enhance their quality. In this paper we introduce our approach for building a repository of heterogeneous alignments. We propose to represent alignments by means of a generic model and we introduce some operators and algorithms to import, compose, intersect and merge alignments. For a concrete application we use ontology alignments and parallel corpora.