Handling effectively big EMF models has often been one of the main barriers that can retain from adopting modeling technologies in very large-scale complex systems. In this talk we present Neo4EMF, a Neo4j-based persistence framework allowing on-demand loading, storage, unloading and actual use of very large EMF models. Neo4EMF provides a No-SQL database persistence framework based on Neo4j, which is a transactional property-graph database that has proved having a remarkable running speed for connected data operations compared to relational databases. In terms of performance, Neo4EMF eases data access and storage not only in a manner to reduce time and memory usage but also to allow big EMF models to fit into a reduced amount of memory. This is made possible through a lightweight and on-demand loading mechanism. Moreover, Neo4EMF comes with a dirty saving mechanism allowing to store huge chunks of data even with limited memory resources.