We have presented a model of a new library organization tool called SOMLib, which is based on a prominent unsupervised neural network architecture, the self-organizing map (SOM). The applicability of SOM and modified architectures to the problem of library organization has been evaluated and shown by many projects so far. SOMLib goes one step further by integrating several independent SOM-based libraries to form higher order library maps. This allows the maintenance of the very libraries locally where training requirements can be met since the maps are relatively small and only a comparably small amount of documents has to be considered. Higher order maps only use the trained lower order maps as basis for their training data to keep both map size and training time low, creating a kind of 'super library'. These higher order libraries do not necessarily need to form a strict hierarchy but can include any other SOMLib system, forming a network of library maps. Users can browse these libraries utilizing the intuitive user interfaces facilitated by the topology preserving mapping of SOMs. Furthermore, documents can be retrieved based on queries, which can either be presented as sample texts, or as keyword queries which in turn can be enhanced by merging the query with a user profile SOM. Due to its modular structure, management of the various libraries is greatly simplified, since they can be independently enlarged and retrained. Finally the system does not pose limitations on architecture variation, so local libraries can be built to best suit the needs of the individual users.