This paper introduces novel supervised learning techniques for real-time selection (triggering) of hadronically decaying tau leptons in proton-proton colliders. By implementing classic machine learning decision trees and advanced deep learning models, such as Multi-Layer Perceptron or residual NN, visible improvements in performance compared to standard tau triggers are observed. We show how such an implementation may lower the current energy thresholds, thus contributing to increasing the sensitivity of searches for new phenomena in proton-proton collisions classified by low-energy tau leptons.