Blog
Token-Oriented Object Notation
Token-Oriented Object Notation
Token-Oriented Object Notation ( TOON ), is a data serialization format designed to be a more compact and LLM – friendly alternative to JSON. It reduces the token count by eliminating redundant syntax like quotes and braces, using indentation for structure and a tabular format for arrays of objects. This makes data cheaper and faster for large language models to process while still being human-readable.
Features of TOON
-
Token efficiency:Â
It minimizes tokens by removing characters like quotes, braces, and commas, which are often repeated in JSON.Â
-
LLM-friendly structure:Â
TOON adds structural metadata that helps LLMs parse and understand the data more reliably, reducing parsing errors.Â
-
Tabular arrays:Â
Uniform arrays of objects, a common data pattern, are represented in a compact, CSV-like tabular format with keys declared once for the entire block.Â
-
Human-readable:Â
The use of indentation and a clean layout makes the data easy for humans to read, similar to YAML.Â
-
JSON compatibility:Â
It’s designed to be a lossless, drop-in replacement for JSON when sending data to LLMs. You can use JSON for your application code and convert it to TOON specifically for LLM prompts.Â
-
Purpose:ÂIt acts as a translation layer between traditional JSON-based systems and LLM prompts, making AI interactions more efficient and cost-effective.Â