Reducing the training time for brain computer interfaces based on steady state evoked potentials, is essential to develop practical applications. We propose to eliminate the training required by the user before using the BCI with a switch-and-train (SAT) framework. Initially the BCI uses a training-free detection algorithm, and once sufficient training data is collected online, the BCI switches to a subject-specific training-based algorithm. Furthermore, the training-based algorithm is continuously re-trained in real-time. The performance of the SAT framework reached that of training-based algorithms for 8 out of 10 subjects after an average of 179 s ±33 s, an overall improvement over the training-free algorithm of 8.06%.