TOON

TOON (Token-Oriented Object Notation) is an open-source, MIT-licensed format for passing structured data to large language models in fewer tokens. The project provides a TypeScript SDK, a CLI, a playground and benchmarks, and the current release is v4.4.0.

What It Is

TOON is a serialization format that encodes the same objects, arrays and primitives as JSON, with deterministic, lossless round-trips. It uses indentation instead of braces and minimizes quoting, which the site describes as YAML-like readability with CSV-style compactness. It is meant for LLM prompts and outputs rather than as a general replacement for JSON in every context.

How the Format Works

Uniform objects, whether in an array or under keys, use a tabular form: the field list is declared once and each object follows as a single row. Array headers declare the row count with [N] and the field names with {fields}, so the site says truncated or malformed output cannot slip through unnoticed. The documentation describes four forms of the format, along with quoting and type rules.

Efficiency Claims

The project's benchmarks page states that TOON matches JSON's retrieval accuracy (72.2% vs 71.4%) while using 42.6% fewer tokens. These are figures published by the project itself.

Tooling and Ecosystem

The documentation covers a getting-started guide, guidance on using TOON with LLMs, a CLI, an API reference, a syntax cheatsheet, a specification and an efficiency formalization. The site says there are official implementations plus dozens of community ports, all targeting one spec with a shared conformance test suite.