Depression is a significant global health issue, with increasing prevalence among various age groups, especially severed by the COVID-19 pandemic. Early detection of depression is crucial for effective intervention because of its traditional methods may be time-consuming. This study proposes a hybrid deep learning model, BiLSTM-CNN, combined with GloVe word embeddings and TF-IDF feature extraction, to detect depression from textual data in the Indonesian language. The dataset used in this study is collected from X consist of 50.523 tweets then manually labeled using a majority vote system. Various scenarios were evaluated, including testing the best split ratios, n-gram, maximum feature, then apply GloVe as feature expansion with three different built corpuses consist of 50.523 data from tweet, 100.594 data from indonews, and the combination of both. The result of this study is that BiLSTM model earned the highest accuracy of 84.2% which has increased by 0.67% from base model because of the model architect