Overview
Kilosort is the premier open-source framework for spike sorting, specifically engineered to process data from high-density electrophysiology probes such as Neuropixels. Now in its fourth major iteration (Kilosort4), the tool has transitioned from a MATLAB-dependent architecture to a robust, pure Python implementation, significantly lowering the barrier for entry while enhancing performance. Kilosort employs a sophisticated template-matching algorithm that iteratively identifies and subtracts neural waveforms from raw binary data. A critical component of its architecture is the integrated drift correction, which utilizes spatial-temporal registration to account for tissue movement relative to the probe—a common challenge in chronic in vivo recordings. By leveraging NVIDIA CUDA for GPU acceleration, Kilosort can process thousands of channels in near real-time, outpacing traditional manual or semi-automated methods. Its 2026 market position remains dominant within the global neuroscience community, serving as the standard backend for high-throughput data pipelines in academic and clinical research institutions. The software is designed to produce outputs compatible with curation tools like Phy, facilitating a seamless transition from raw signal to publishable single-unit activity data.