What is this?
TOON (Token-Oriented Object Notation) is a compact, human-readable way to serialize JSON data so it spends fewer tokens in LLM prompts. The repo ships a TypeScript SDK, a CLI, and benchmarks.
Is TOON a file format, a library, or a command-line tool?
All three layers: TOON defines the text format itself, a TypeScript SDK encodes and decodes it in code, and a CLI converts JSON from the terminal.
How do I start using it?
Install @toon-format/toon from npm and call the SDK to encode your JSON before sending it to a model, or use the CLI to convert JSON files from the terminal. The source is a pnpm monorepo: pnpm install, then pnpm build.
Do I need anything besides Node to run it?
No — the SDK and CLI are pure TypeScript with no Python, Go, or Rust components; pnpm is only needed if you build the monorepo from source.
Will it always use fewer tokens than JSON?
Not guaranteed for every payload — the project publishes benchmarks, so measure the savings on your own data shape before rewiring your prompts.
Is TOON human-readable, or is it another binary format?
It is plain text by design — the project describes it as human-readable, so you can inspect exactly what you are about to send to the model before sending it.
Can languages other than TypeScript work with TOON?
This repository provides the TypeScript SDK and CLI, but TOON itself is just a text notation, so any language that can emit text can implement it.
How popular is the project?
Very — since it appeared in October 2025 it has gathered over 25,000 stars on GitHub.
What are similar projects?
headroom, protobuf.js, billion-context, liteparse, dsh-plugin-agent-workflow, fast-jev-compaction, caveman, codeburn
Which topics is it in?
tokenization, serialization, token-usage, prompt-engineering, json, sdk, benchmark, npm, terminal, tool, llm, nodejs