An experimental trainable text codec for developers who already know their data — MQTT payloads, telemetry, logs, JSON-like records, protocol text.
Instead of trying to be a universal compressor, loxc lets you train a codec table from representative sample data, export that table, transfer it, and load it later in an application, tool or embedded runtime.
Symbols chosen by frequency in your training data.
Matrix-based symbol layout with nested submatrices for less frequent data.
Binary encoded output with trained lookup tables.
Predictable table-driven decoding.
A small C99 integration surface, consistent with the rest of the LOX family.
Train on the host, export, load on the device.
loxc is not intended to compete with gzip, zstd, brotli or lz4 — those are mature general-purpose compressors. It is also not encryption and not a universal archive format. Its value is narrower: a compact table-driven encoding path for specific text payloads.