We developed a novel free-running data acquisition system for the AMBER experiment. The system features a hybrid architecture containing a scalable FPGA-based system for data collection and conventional distributed computing for data reduction. The current implementation can collect up to 10 GB/s sustained data rate. The FPGA system substitutes high-performance networks by merging time-correlated data and distribution between computers. The data reduction is performed by a filtering farm decreasing the incoming data rate by a factor of 50 to 100-200 MB/s. The filtering framework implements various data reduction algorithms for different physics programs. These algorithms perform partial data decoding, time, and spatial analysis of the data in order to select predefined event topology in a semi-online manner. Our system also performs continuous and iterative time calibration of detectors, which is required by the continuously running acquisition system. Additionally, we developed a simulation tool able to emulate detector responses to particles passing the AMBER spectrometer and convert them into correctly formatted raw data. These generated data are used to test and validate the readout chain and the filtering framework. The entire system will be tested with a limited number of detectors this year. The first physics run is planned for 2024.