Cloud-based applications are problematic from a privacy perspective because they typically have access to large amounts of user data and metadata. This centralisation of user data creates an attractive target for actors such as criminals, suppressive governments, and companies selling the data. At the same time, the popularity of mobile and web applications has led to a growing amount of sensitive data being stored in the cloud. This dissertation focuses on collaborative applications, such as Google Docs and Microsoft Office Online, where users currently rely on cloud-based solutions. It explores decentralised alternatives that allow the use of end-to-end encryption and anonymous communication systems to improve both information privacy and communication privacy. One approach for a collaborative application to synchronise data in a privacy-preserving way is to use Tor hidden services, providing end-to-end encrypted communication, while also hiding collaborators’ identity. However, running Tor comes at a cost. We explore the costs of running a hidden service on a smartphone. Smartphones are nowadays the most frequently used computing devices, but they are also relatively resource-constrained. We build an empirical model of monthly cellular data traffic, and estimate a median 198 MiB for a typical user. We further estimate that the network activity would cost at least 9.6% of daily battery capacity on a Nexus One using 3G Internet. We explore four optimisations that, in combination, reduce the estimated median data cost to 61 MiB. We also consider the security and privacy properties of decentralised collaborative applications, and explore a challenge that is introduced by a decentralised design – the lack of a trusted server guaranteeing consistency between collaborators. We present a novel snapshot protocol that ensures consistency, whilst allowing the past edit history to be hidden from new collaborators, and without relying on a consensus mechanism. Lastly, we evaluate the overhead of the snapshot protocol by replaying editing histories from 270 Wikipedia articles, and demonstrate how its correctness and security properties are achieved. Assuming the number of collaborators remains small, the protocol is scalable in terms of CPU, memory, and network usage. It substantially reduces the amount of data transferred to a new collaborator compared to a basic protocol that transmits the full history. The computational cost is in the order of milliseconds per operation, indicating the protocol is suitable for applications where the rate of edits is relatively low.