This paper provides an overview of MediaDART, a framework for building online services for distributing and sharing digital media. Inspired by the participative model of Web 2.0, MediaDART relies on a scalable and decentralized architecture that can grow with the contribution of users. The architecture is based on an arbitrary number of nodes interconnected through a p2p network implementing a distributed hash table (DHT). The DHT provides resource storage and parallel resource processing for operations of feature extraction, adaptation and composition. MediaDART adopts application-level multicast based on distribution trees for delivery in streaming and implements algorithms to dynamically replicate resources across the network. The framework allows content description through user-defined tags. Tools for personalized content retrieval based on recommendation algorithms and user profiling are included too. This paper also describes two prototype applications and outlines further work.