The Numdam French digital mathematics library has now been in operation for more than 15 years with no major upgrade. It holds more than 57000 documents either scanned or born digital (spanning over one million pages). The information system has been recently completely redesigned. In this article, we present the new Numdam ecosystem. A metadata factory is used to store metadata from a variety of sources, normalize it under JATS (articles) or BITS (books) XML formats, and enhance it through manual editing or automated agents (tagging math formulas, matching to external databases and interlinking, etc.). The data model supports the main types of documents currently expected to populate the DML: journals, seminars, conference proceedings, multivolume works, books, book parts, doctoral theses. All documents are in collections that can belong to one or more corpus. The workflow has been simplified and allows easy deployment on test and production web sites. A platform holds all the data in one place, can generate multiple web sites, each with a different view on the data, and provides an OAI-PMH server to the outside world. Finally, the article presents future plans to create a DML-ready platform based on the new Numdam platform.