Bolt is a discriminative learning of linear predictors (e.g. SVM or Logistic Regression) that uses fast online learning algorithms to aimed large-scale, high-dimensional and sparse machine-learning problems. In particular, problems encountered in information retrieval and natural language processing.