Convert JSON to TOON only when it really saves tokens
datoon evaluates payload shape, depth, and expected savings before converting. It keeps raw JSON when TOON is not a net win, and returns a transparent conversion report for every run.
01
Why teams use datoon
Built for practical AI/data pipelines where prompt size, readability, and deterministic behavior all matter.
Decision
Decision-first conversion
Auto mode checks uniform tabular structure, payload depth, and minimum savings before converting to TOON.
Reporting
Transparent reporting
Every run emits a report with token estimates, conversion decision, reason, and savings ratio.
Integration
CLI and plugin ready
Use as a Python CLI in CI/pipelines or install as Claude Code plugin via marketplace.
02
Benchmark snapshot
Local suite compares JSON baseline, forced TOON conversion, and datoon auto mode across five reference payloads.
| Metric | Value |
|---|---|
| Forced TOON avg reduction | 26.8% |
| Auto mode avg reduction | 28.1% |
| Auto decisions | 3 / 5 payloads converted |
| Failed forced conversions | 0 / 5 |
Reproduce locally with
PYTHONPATH=src python benchmarks/run.py.
03
Install and run
uv sync
# stdin usage
echo '{"users":[{"id":1,"name":"Ada"}]}' | datoon --report-stdout
# file usage
datoon ./input.json -o ./output.toon --report ./report.json
Requires Python 3.12+ and Node.js for TOON CLI
execution through
npx --yes @toon-format/cli@2.