Profile any CSV without uploading it.
Types, nulls, distinct counts, quantiles, histograms and data-quality warnings for every column, in one streaming pass. This page runs the same Zig engine as the command-line tool, compiled to a 256 KB WebAssembly module. Your file is read in slices and never leaves this tab.
| # | column | type | nulls | distinct | min | median | max | shape |
|---|
Click a column for quantiles, top values and details. ~ marks a sketch estimate.
One pass, bounded memory
Each column keeps fixed-size streaming summaries rather than its values: HyperLogLog for distinct counts, a KLL sketch for quantiles and histograms, Misra-Gries for the most common values, and Welford for mean and spread. Small columns stay exact; large ones switch to sketches with known error.
Streams from disk, even here
A Web Worker hands the WebAssembly module one slice of your file at a time through FileReaderSync,
so a multi-gigabyte export is profiled with a few megabytes of memory. Gzip is decompressed on the fly.
Same engine as the CLI
On the command line it streams at about 290 MiB/s on all cores and can fail a CI job when data looks wrong:
parakh --fail-on-warn export.csv.gz