A data processing pipeline for serial femtosecond crystallography at SACLA was developed, based on Cheetah [Barty et al. (2014). J. Appl. Cryst. 47, 1118-1131] and CrystFEL [White et al. (2016). J. Appl. Cryst. 49, 680-689]. The original programs were adapted for data acquisition through the SACLA API, thread and inter-node parallelization, and efficient image handling. The pipeline consists of two stages: The first, online stage can analyse all images in real time, with a latency of less than a few seconds, to provide feedback on hit rate and detector saturation. The second, offline stage converts hit images into HDF5 files and runs CrystFEL for indexing and integration. The size of the filtered compressed output is comparable to that of a synchrotron data set. The pipeline enables real-time feedback and rapid structure solution during beamtime.