RGS aims to be an R package for flexibly defining and fitting hierarchical Bayesian models by eventually defining custom samplers in either C or R, and obtaining simulators with good performances (e.g., compared with JAGS). The typical RGS workflow would be the following:

1. specify the hierarchical model as a Directed Acyclic Graph (DAG), like in OpenBUGS and JAGS
2. customly assigns samplers, or 'rules to update', to each model node
3. start the MCMC loop

A key target of the package is obtaining decent performances compared with other engines, so that the package is usable in real world problems and for true experimentation.

Package vignette can be seen here.