Datadog's Auto Smoother is a smoothing function that helps identify trends in noisy timeseries data, particularly in high-scale infrastructure and applications. It automatically chooses the optimal window size to smooth the timeseries, adapting to changes in noise levels as new data is collected. This algorithm uses a combination of roughness and kurtosis measures to ensure that the smoothed series preserves large-scale trends while preventing oversmoothing. Auto Smoother provides several advantages over traditional smoothing functions, including automatic parameter selection, real-time adaptation, and consistency across different time ranges and hosts. It can be used in conjunction with other algorithms to highlight important abnormalities in metrics, making it easier to extract valuable signals from noisy timeseries data.