Overview
SpikeSift is a high-performance spike sorting algorithm designed for high-density extracellular recordings. It delivers state-of-the-art accuracy while running in real time on a single CPU core.
How Does SpikeSift Work?
SpikeSift extracts and tracks individual neuron activity from raw extracellular recordings. It includes:
Filtering and segmenting the signal based on drift-aware heuristics
Iteratively detecting and clustering spikes within each segment
Merging matching clusters across segments to produce globally aligned spike trains
The result is a drift-corrected, neuron-by-neuron reconstruction of spiking activity across time.
Why Use SpikeSift?
SpikeSift is built for speed, robustness, and clean integration into any workflow:
Extremely fast — sorts thousands of channels in real time on a single CPU core
Drift-resilient — handles both gradual and abrupt electrode drift
Clean and non-intrusive — no data copying, no file modifications, no clutter
Modular — sort in parallel, split or merge segments, track transients
Drop-in ready — works out of the box on most datasets
Session-aware — merge across files, append sessions, or sort progressively
Reliable on short recordings — maintains accuracy even with limited data
These features make SpikeSift ideal for real-time pipelines, high-throughput labs, and large-scale sorting tasks — even on resource-constrained systems.
What the Documentation Covers
The rest of this documentation includes:
Installation — how to install and get started
User Guide — how to sort data and interpret the results
Example Usage — real-world workflows and advanced use cases
Performance — benchmarks and key efficiency advantages
Implementation — algorithmic insights and design principles
API Reference — complete API for all user-facing functions
For a quick start, see the User Guide. To explore practical workflows, head to Example Usage.
Note
SpikeSift is under active development and continues to improve in accuracy and flexibility. For more details or citations, see the upcoming preprint: