Quantile regression forests (QRF) is a non-parametric, tree-based ensemble method for estimating conditional quantiles. It is a generalization of the random forests algorithm, which has proven extremely popular and useful as a general-purpose machine learning method. Unlike random forests, which output the weighted mean value of training labels, QRF uses the weighted empirical distribution of training labels to obtain the predictive distribution, enabling probabilistic predictions for regression problems. The quantile-forest package provides a fast, feature-rich QRF implementation optimized with Cython for training and inference speed. It allows estimating arbitrary quantiles at prediction time without retraining and includes methods for out-of-bag estimation, quantile rank calculation, and proximity counts. The package is compatible with and can serve as a drop-in replacement for scikit-learn's forest regressors. It has been cited in scholarly work and used in production settings, including at Zillow. QRF is useful for understanding relationships between variables outside the mean, particularly for non-normally distributed outcomes or nonlinear relationships. It provides predictions for various quantiles, allowing researchers to quantify uncertainties and capture the full spectrum of potential outcomes. QRF has become a standard method for probabilistic prediction in machine learning. Traditional prediction intervals often rely on assumptions such as normality, but QRF allows non-parametric, flexible, and adaptive prediction intervals. The package provides utilities like out-of-bag scoring, quantile rank calculation, and proximity estimation, enhancing its applicability for researchers and practitioners. The package enables researchers to estimate conditional quantiles accurately, providing a powerful tool for quantile regression and uncertainty estimation.Quantile regression forests (QRF) is a non-parametric, tree-based ensemble method for estimating conditional quantiles. It is a generalization of the random forests algorithm, which has proven extremely popular and useful as a general-purpose machine learning method. Unlike random forests, which output the weighted mean value of training labels, QRF uses the weighted empirical distribution of training labels to obtain the predictive distribution, enabling probabilistic predictions for regression problems. The quantile-forest package provides a fast, feature-rich QRF implementation optimized with Cython for training and inference speed. It allows estimating arbitrary quantiles at prediction time without retraining and includes methods for out-of-bag estimation, quantile rank calculation, and proximity counts. The package is compatible with and can serve as a drop-in replacement for scikit-learn's forest regressors. It has been cited in scholarly work and used in production settings, including at Zillow. QRF is useful for understanding relationships between variables outside the mean, particularly for non-normally distributed outcomes or nonlinear relationships. It provides predictions for various quantiles, allowing researchers to quantify uncertainties and capture the full spectrum of potential outcomes. QRF has become a standard method for probabilistic prediction in machine learning. Traditional prediction intervals often rely on assumptions such as normality, but QRF allows non-parametric, flexible, and adaptive prediction intervals. The package provides utilities like out-of-bag scoring, quantile rank calculation, and proximity estimation, enhancing its applicability for researchers and practitioners. The package enables researchers to estimate conditional quantiles accurately, providing a powerful tool for quantile regression and uncertainty estimation.