<p>Streaming data is generated at high speed and in large quantities over time, and it calls for online learning to deal with it. In this paper, a new online updating method is established for smoothing quantile regression to make inferences in real-time. The renewable estimators are updated only by the current dataset and summary statistics of historical datasets. This method is adapted to the streaming datasets containing small samples. Theoretically, it is proved that renewable estimators have consistency and asymptotic normality and equivalence to pooled offline estimators based on all datasets. The dynamic bandwidth selection is applied to estimate the asymptotic covariance matrix in an online manner, which is theoretically highly asymptotically efficient. In particular, the renewable estimator provides asymptotic confidence intervals that are asymptotically smaller than those generated by existing methods, thereby improving the accuracy of interval estimation. Additionally, our approach addresses the common assumption of homogeneous models by accommodating non-parametric heterogeneity and detecting and removing anomalous data batches through an online screening process. Meanwhile, numerical simulations verify the theoretical results and outcomes on real datasets illustrate that our method is adapted to real streaming data situations.</p>