Our paper discusses simulation-based Bayesian inference using information from previous draws to build the proposals. The aim is to produce samplers that are easy to implement, that explore the target distribution effectively, and that are computationally efficient and mix well.