CLI

dspk from the command line: pipelines, devices, recordings, detection, sorting, benchmarks, streaming, model hub

dspk [OPTIONS] <COMMAND>

Command	What it does
run	Run a pipeline spec (.json or .toml) over its inputs, or inputs given here
spec	Check a pipeline spec, or print the spec’s JSON Schema
doctor	What this build can do here: version, features, runtimes and their devices
autotune	Tune this machine’s devices ahead of time, so first runs don’t stop to benchmark
inspect	Launch geometries the selected runtime gives for a channels × samples buffer
open	Open a recording (any format dspk-io reads) and report its layout and read speed
probe	Show a probe layout preset
detect	Detect spikes in a whole recording and write them as a sorting (one unit per channel)
sort	Spike-sort a recording (Kilosort4, EMUsort, MountainSort 5, SpyKING CIRCUS 2, Tridesclous 2) with the separate dspk-cli-sorter program (GPL-3.0; dspk sorters install)
sorters	The dspk-cli-sorter program sort runs: install it, or show which one is used
convert	Convert a sorting between formats (Phy, .sorting.zarr, NWB units)
generate	Write a synthetic recording (noise, line noise, drifting units)
benchmark	Time processing on the selected runtime
serve	Serve a recording over QUIC (signal and views)
receive	Receive a served recording: throughput, latency, optional save
clients	Who may connect to serve: enroll, revoke, list clients (certificates from the server’s own authority)
remote	Work with a serve server: info, files, sort on it, status, logs, stop
hub	Published model and sorter artifacts: list, info, pull, verify, remove, clean

Options of every command

• --runtime <RUNTIME>: env DSPK_RUNTIME
  Compute runtime: wgpu, cpu, cuda or hip, among those compiled in (default: the first compiled in, GPUs first; doctor shows which work on this machine)
• -v, --verbose
  Show debug logs (runtimes, transport)
• -h, --help
  Print help
• -V, --version
  Print version