We introduce a system-level architecture providing fine-grained control over user privacy, in the context of location-based services accessed via mobile devices. In contrast with most mobile platforms today, users only have coarse-grained control over their privacy, either accepting to unconditionally stream their locations in order to use a service, or renouncing the service altogether. However, not all location-based services do require the same level of location accuracy and the same level of privacy renouncement. With this architecture, the user can adapt the tradeoff between location privacy and location accuracy. To achieve this, our architecture relies on three main elements: a trusted module extending the underlying mobile platform, a secure protocol between that module and untrusted applications offering location-based services, and a tree capturing user's zones of interest and organizing them in various accuracy levels. Untrusted mobile applications no longer receive user locations directly: the trusted module intercepts them to compute user's zones of interest and create the tree. The user can then decide what level of accuracy will be disclosed to what application. We evaluate this architecture from a privacy preserving point of view by comparing well-known blurring mechanisms and our tree.