soon-format

SOON (Shape-Oriented Object Notation) — lossless, token-efficient encoding of nested JSON for LLM prompts. Python implementation.

Install

pip install soon-format            # library + `soon` CLI
pip install "soon-format[tokens]"  # + tiktoken-based cost decisions

Usage

from soon_format import encode, decode, stats

doc = encode(data)                        # auto mode, never-worse guarantee
data == decode(doc)                       # True — lossless
stats(data)                               # {"json_chars": …, "soon_chars": …, "saving": 0.604}

# Real-token cost decisions
encode(data, tokenizer="o200k_base")
stats(data, tokenizer="o200k_base")       # adds json_tokens / soon_tokens

CLI

soon encode data.json --stats
soon decode doc.soon              # compact JSON to stdout
soon decode doc.soon --pretty     # indented JSON
soon check data.json              # round-trip verification

Spec & conformance

This implementation passes the shared conformance suite byte-for-byte. See the SOON spec for the normative format definition.

License

MIT © Yasin Ughur