Analyze at multiple levels
Count individual bytes, recurring sequences, and structural word patterns in the uploaded file.
Lossless file compression
Find the structure a byte-by-byte encoder misses.
ByteSize combines multi-level frequency analysis, a bit-cost decision rule, and canonical Huffman coding to reduce repetitive files without changing a single original byte.
Server-side processing · SHA-256 verified · No lossy conversion
NEW COMPRESSIONChoose a file
Balanced
Drop a file here
Accepted inputs include text, source code, markup, structured data, logs, and PDF.
MODEAdaptive
VERIFYSHA-256
BACKENDC++ native
The method
One pipeline. Three deliberate decisions.
The system stays inside the Huffman family. It does not switch to an unrelated compressor behind the scenes.
Keep only patterns that pay
The Bit Cost Decision Engine rejects candidates whose dictionary cost would erase their savings.
Encode and verify
Canonical Hybrid Huffman coding produces the AFC file, then an automatic round trip confirms exact bytes and SHA-256.
Where the method applies
Built for files that still have structure to find.
Every category below is one ByteSize has been measured on, and each result is published in this repository. PDF depends entirely on what is inside it: a book stored as text streams reduces like the text it contains — 71.8% on the Silesia reymont volume — while a PDF of scanned images has nothing left to find, and across 200 real-world documents the average was 12.5%. Declared in-scope file types are accepted for new compression, and every accepted file is reconstructed byte for byte.
TXTCSVJSONSQL
XMLHTMLLOG
Source codePDF
Ready when you are