TOON: Token-Oriented Object Notation

What is TOON?

TOON (Token-Oriented Object Notation) is a compact data format designed to minimize token usage when communicating with Large Language Models (LLMs). It reduces API costs by 30-60% compared to JSON while maintaining human readability.

Why TOON Matters

LLMs like GPT, Gemini, and Claude charge per token. TOON optimizes data transfer by:

• Stripping unnecessary punctuation (braces, quotes, commas)
• Using single headers for repeated data structures
• Employing CSV-like tabular layouts for arrays

Core Syntax

Objects

user:
  name: John Doe
  age: 30
  email: john.doe@example.com
        

Arrays

users[2]{id,name,email}:
1,Alice,alice@example.com
2,Bob,bob@example.com
        

Nested Data

company:
  name: TechCorp
  departments[2]{name,budget}:
  Engineering,500000
  Marketing,200000
        

JSON vs TOON Comparison

JSON (85 tokens):

{
  "company": {
    "name": "TechCorp",
    "departments": [
      {"name": "Engineering", "budget": 500000},
      {"name": "Marketing", "budget": 200000}
    ]
  }
}
        

TOON (40 tokens - 53% savings):

company:
  name: TechCorp
  departments[2]{name,budget}:
  Engineering,500000
  Marketing,200000
        

Key Benefits

• Cost Reduction: 30-60% fewer tokens = lower API bills
• Better Accuracy: Explicit schema markers reduce LLM parsing errors
• Larger Context: Fit more data in limited context windows
• Faster Processing: Compact format parses quicker

Implementation

TOON works as a translation layer:

1. Use JSON internally and for databases
2. Convert to TOON before LLM API calls
3. Convert LLM responses back to JSON

Libraries:

• JavaScript/TypeScript: toon-js
• Python: pytoon
• Java: TOON4J

Basic Usage:

const toonData = jsonToToon(jsonInput);
const llmResponse = await callLLM(toonData);
const result = toonToJson(llmResponse);
        

Use Cases

• AI Chatbots: Longer conversation histories
• Data Analytics: Efficient dataset processing
• API Integration: Optimized microservice communication
• Real-time Applications: Faster data exchange

Best Practices

• Start with high-impact, data-heavy API calls
• Monitor token usage and cost savings
• Use consistent schemas with array length markers [N]
• Implement error handling with JSON fallbacks

Limitations

• Learning curve for new syntax
• Not ideal for simple, one-off data
• Limited ecosystem (growing rapidly)
• Not a JSON replacement for all use cases

Getting Started

1. Identify token-heavy LLM interactions
2. Choose a TOON library for your language
3. Convert critical data paths to TOON
4. Monitor performance and cost improvements

TOON represents the future of AI-optimized data communication, making LLM applications more efficient and cost-effective.