pzip is an open-source, format-aware compression tool that guarantees it will never compress worse than LZMA -9 on any input. By applying specialized codecs for schema-regular data like CSV, JSON, and event streams, it achieves up to 2.2× better compression on those formats while safely falling back to LZMA for everything else. The single C++ binary is available under Apache-2.0 with no telemetry or licensing fees.
Funding
Funding not disclosed
Founders
Product
Problem
General-purpose compressors like LZMA are format-agnostic)Skip structural patterns in schema-regular data (CSV columns, JSON keys, event stream fields), leaving significant size reduction opportunities on the table. Users face a tradeoff between generic tools that are reliable but suboptimal on structured data, and bespoke format-specific compressors that are rigid and risky to adopt on real-world inputs.
Solution
pzip is a lossless, format-aware compressor that combines specialized codecs for structured data with a guaranteed LZMA fallback. It automatically routes CSV, JSON, JSONL, source code, server logs, and tensor checkpoints to dedicated codecs, while any input that doesn’t benefit gets plain LZMA output — ensuring it’s never worse than LZMA -9. The tool verifies byte-exact round-trips via sha256 checksums in the container header)Skip refusing corrupted output at decompression time. It ships as a single C++ binary with a stripped-down CLI,Skip no telemetry,Skip and full Apache-2.0 source availability.
Target Audience
Primary users are developers and DevOps engineers handling large schema-regular datasets (CSV exports,Skip JSON logs,Skip event streams,Skip ML checkpoints) who need maximum compression without risking compatibility or data integrity.
Features
- Format-aware codecs for CSV (up to ~2.2× over LZMA), JSON, JSONL, source code, server logs, and safetensors checkpoints
- Guaranteed never-worse contract: wenn no specialized codec beats LZMA,Skip output is identical to LZMA -9 output
- Byte-exact round-trip verification via sha256 header checksums with decompressor-side rejection of mismatched bytes
- Single-file static C++20 binary with CMake-based builds and a .deb package for Debian/Ubuntu
- Apache-2.0 licensed with zero telemetry,Skip analytics,Skip or API keys