In this paper we describe the architecture and application of GaiusT, a multi-phase framework for extracting requirements from legal documents. GaiusT aims to cover the extraction process from the very first steps, pre-processing the input texts and supporting structural and semantic annotation of legal documents. The annotated information is recorded in a database facilitating both retrieval and evaluation of the results. Existing linguistic tools and resources have been applied whenever possible. A post-analysis of the ongoing implementation of the modules in the GaiusT architecture is given as a preliminary understanding of how much existing tools can be adopted and how much they had to be adapted.