ConvTimeNet is a novel deep hierarchical fully convolutional network designed for multivariate time series analysis. The key design of ConvTimeNet addresses two main limitations of traditional convolutional networks: it proposes an adaptive segmentation of time series into sub-series level patches, treating these as fundamental modeling units, and it integrates deepwise and pointwise convolution operations to effectively capture both global sequence and cross-variable dependencies. The network's backbone is inspired by Transformer encoders, allowing for multi-scale representations and efficient computation. Extensive experiments on time series forecasting and classification tasks demonstrate that ConvTimeNet outperforms strong baselines, including advanced Transformer networks and pioneering convolutional models. The code for ConvTimeNet is publicly available.ConvTimeNet is a novel deep hierarchical fully convolutional network designed for multivariate time series analysis. The key design of ConvTimeNet addresses two main limitations of traditional convolutional networks: it proposes an adaptive segmentation of time series into sub-series level patches, treating these as fundamental modeling units, and it integrates deepwise and pointwise convolution operations to effectively capture both global sequence and cross-variable dependencies. The network's backbone is inspired by Transformer encoders, allowing for multi-scale representations and efficient computation. Extensive experiments on time series forecasting and classification tasks demonstrate that ConvTimeNet outperforms strong baselines, including advanced Transformer networks and pioneering convolutional models. The code for ConvTimeNet is publicly available.