We present the Democrat project, "Description, modelling and automatic detection of coreference chains in French," and its four objectives, that is to provide: (i) an integrated, discursive, diachronic and inter-genre description of coreference chains; (ii) a corpus of written French texts with annotated coreference chains; (iii) several tools for visualizing and exploring the coreference chains; (iv) two NLP systems that are able to process raw text written in French and to extract referring expressions as well as coreference chainswhich have also brought innovations to the field of deep learning. We present the main results of Democrat and we describe the work steps that made it possible to obtain them, in particular the corpus, which was manually annotated by forty members of the project.