TOON is a modern data format designed specifically for Large Language Models. It represents the same objects, arrays, and values you have in JSON, but with a much more compact, table-like syntax. Instead of repeating keys and punctuation over and over, TOON declares the structure once, then lists the data row by row.
For uniform data (for example, an array of products, users, or orders), TOON can often use 30–60% fewer tokens than JSON while remaining human-readable and easy for models to parse.
JSON (JavaScript Object Notation) is the universal
language of APIs and web apps. It’s supported by every programming
language, it’s easy to log, easy to debug, and perfect when machines
talk to machines. The downside, especially in AI workloads, is that
JSON is very verbose: every {}, [],
" and repeated key adds extra tokens and extra cost in
LLM prompts.
Think of TOON as a prompt-friendly view of your JSON. You can keep JSON in your code and databases, and only convert to TOON when talking to an LLM – exactly what this tool does.