This paper introduces and analyzes a new model for realized volatility that accommodates endogenous regime switching. The model is based on the heterogeneous autoregressive model and allows for two-state regime switching. Importantly, a current shock to the realized volatility affects the regime switching in the next period. We apply the model to the realized volatility of the daily S$&$P 500 Index return series. The estimation result shows that the short-term volatility component is the most influential in the high volatility regime while the long-term volatility component is dominant in the low volatility regime. Moreover, the model significantly outperforms existing models in within-sample fitting.