toon-cli 0.1.2

A production-grade CLI for converting JSON to TOON and back.

toon-cli

toon-cli is a production-grade Python CLI for converting JSON into TOON (Token-Oriented Object Notation) and back. TOON is designed to stay human-readable while reducing tokenizer cost for LLM workflows through indentation-based structure and repeated-key compression.

Why TOON

TOON aims to preserve JSON semantics while cutting token overhead:

• Lossless round-trips between JSON and TOON
• Compact table notation for arrays of similarly shaped objects
• Token-aware benchmarking using tiktoken
• Strategy comparison against minified JSON baselines
• Path-level token attribution to show where payload weight lives
• Optional lossy transforms for prompt-oriented payload trimming
• Readable, diff-friendly text for prompt and context pipelines

Example JSON:

{
  "users": [
    {
      "id": 1,
      "name": "Alice",
      "role": "admin"
    },
    {
      "id": 2,
      "name": "Bob",
      "role": "user"
    }
  ]
}

Example TOON:

users[2]{id,name,role}:
  1,Alice,admin
  2,Bob,user

Installation

From PyPI:

pip install toon-cli

After installation, confirm the command is available:

toon --help

CLI

toon encode input.json
toon decode input.toon
toon benchmark input.json
toon optimize input.json
toon validate input.toon
toon stats input.json

Common options:

• --output / -o write to a file instead of stdout
• --pretty / --minified choose human-readable or compact output
• --model select a tokenizer model for token counting
• --alias-keys replace repeated keys with short aliases in TOON output
• --sparse-tables compress arrays of irregular scalar objects into table form
• --drop-nulls, --drop-empty, --drop-keys, --max-string-length, and --top-k apply optional lossy transforms during optimization

Examples

Encode JSON into TOON:

toon encode samples/users.json --output samples/users.toon

Decode TOON back into JSON:

toon decode samples/users.toon --pretty

Benchmark savings:

toon benchmark samples/users.json

Pick the lowest-token strategy and emit it:

toon optimize samples/catalog.json --alias-keys --drop-nulls --drop-empty

Validate a TOON file:

toon validate samples/users.toon

Inspect stats:

toon stats samples/catalog.json

Safety Notes

toon-cli is a local file-processing tool. In the current codebase:

• It reads local JSON or TOON files that you choose
• It writes output only to the path you pass or to predictable default filenames beside the input
• It does not send your file contents over the network
• It does not execute code from the input files it reads
• It rejects malformed TOON syntax instead of guessing
• It refuses to overwrite the input file when toon optimize --output points to that same file

Use the lossy optimization flags carefully, because they intentionally modify data:

• --drop-nulls
• --drop-empty
• --drop-keys
• --top-k
• --max-string-length

These are useful for prompt compression, but not for archival or exact data preservation.

TOON Syntax Overview

TOON supports all JSON types:

• Strings
• Numbers
• Booleans
• Null
• Arrays
• Objects

Core syntax:

name: Alice
active: true
count: 3
meta:
  created_by: system
tags[3]:
  - alpha
  - beta
  - gamma

Compact repeated-key compression:

users[2]{id,name,role}:
  1,Alice,admin
  2,Bob,user

Quoted keys and strings use JSON escaping rules when needed.

Project Layout

toon/
|- benchmark/
|- cli/
|- parser/
|- serializer/
|- tokenizer/
`- utils/

Developer Commands

pytest
ruff check .

Packaging

The project is ready for PyPI publishing through standard build or hatch workflows:

python -m build

Design Notes

• Encoder and decoder are fully typed
• Parser validates malformed indentation, headers, and row widths
• Schema inference is used to discover table-friendly arrays of objects
• Streaming parsing is supported from line iterables
• Token-aware optimization can choose compact table layouts when they improve tokenizer efficiency