This paper presents some aspects of the Silfide server, a system dedicated to the delivery of linguistic resources on the web. After presenting the main issues behind the design of such a system, we focus on the editorial choices related to the use of the Text Encoding Initiative to represent our textual documents. In particular, we focus on the accommodations we have had to carry with regards to the TEI header and address the trade-off between extensive enrichment and genericity of the primary data when one wants to precisely mark-up a given document content. As a whole, we show how essential the TEI has proven to be for a project such as ours both from a practical and conceptual point of view.