Supported data formats are listed below.

Extension	Implied format	Details
.csv	Comma separated values.	see below…
.pxt + .pxb	Pixie fast lightweight binary format.	see below…

CSV files

Rules for CSV files are as follows:

1 header row.
- Header row values are comma-separated and will be interpreted as the timeseries names.
Timestamps should be in the first column. The column name for this column doesn’t matter. Timestamps are allowed to take a range of valid ISO 8601 formats. Some good choices are:
- yyyy-MM-dd e.g. 2022-03-21 for daily data.
- yyyy-MM-dd'T'HH:mm:ss e.g. 2019-06-28T22:00:00 for sub-daily data. Note that the Python formatter for this is “%Y-%m-%dT%H:%M:%S”.
Timesteps in the file must be regular (no missing timestamps), and in temporal order.
Blank values (or whitespace) are taken to indicate missing data.

Pixie binary files (.pxt + .pxb)

The Pixie binary format stores one or more timeseries efficiently using a compression algorithm inspired by Gorilla. It always uses two paired files that share the same base name. Both files are required.

.pxt — index file. A plain-text CSV with one row per series, listing the series name, byte offset into the .pxb file, start/end times, timestep, and length. This file is human-readable and can be useful for inspecting the contents of the dataset.
.pxb — binary file. Contains the actual compressed timeseries data. Each series is written as a single block. Binary offsets recorded in the .pxt file allow individual series to be read without decompressing the whole file.

Advantages

Smaller files. Pixie files are typically 5–10× smaller than the CSV equivalent. The savings are bigger when timeseries are longer and more regular.

Commandline output flag	Files written	Total size
-o results.csv	results.csv	263 MB (72 MB if zipped using aggressive zip compression)
-o results.pxb	results.pxt + results.pxb	39 MB
-o results.pxt	results.pxt + results.pxb	39 MB

Random access. Individual series can be loaded without reading the whole file. This makes access way faster than CSV.
Fast encoding and decoding algorithm. Encoding or decoding happens in one sweep using operations that are very cheap for the cpu. You wont notice it happen.
Lossless. Round-tripping a series through .pxt/.pxb returns the original values exactly, including special values like NaN.

Disadvantages

You cannot see the data in a text editor. Special tooling is needed to read the data. The index file partly mitigates this downside.