This report presents the Garcinia peer-to-peer distributed computing framework. Garcinia is implemented on the top of Java Platform and uses IP multicasting to create an unstructured overlay network layer. The framework consists of the agent and the shared library. The agent acts as a node in the peer-to-peer network and also as worker and dispatcher service provider to the consumer of the shared library. The agent keeps an up-to-date list of neighbouring nodes information, provides that information to the requesting consumer application, and accepts the task submitted by the consumer application. All communication between the agent and the consumer application is done via the shared library. A local agent instance must be started to ensure the fairness of the network. The framework provides an interface for developing application to solve bag-oftasks kind of processing. The framework gives the flexibility to the consumer application to choose the worker or to use the built-in mechanism to perform randomised dynamic load balancing. The randomised dynamic load balancing in this framework can attain 62.2% system efficiency compared to ideal parallel ensemble.