The Golden Rule

Use TOON when you have structured, tabular data for LLMs.

If your data is an array of objects with the same schema, and you're feeding it to an AI model, TOON will save you 30-60% in tokens and costs.

Perfect Use Cases

RAG Systems (Retrieval-Augmented Generation)

RAG injects database records into LLM context. TOON massively reduces token usage.

Example Scenario

Your chatbot retrieves 50 relevant products from your database to answer "What laptops are under $1000?"

JSON Approach

~2,100 tokens

TOON Approach

~920 tokens (-56%)

Why TOON wins: RAG queries happen thousands of times daily. Every token saved = massive cost reduction at scale.

Database Exports for AI Analysis

Sending SQL query results to GPT for analysis, insights, or report generation.

Example Query

SELECT id, name, revenue, region, employees FROM companies WHERE revenue > 1000000 ORDER BY revenue DESC LIMIT 500;

Why TOON wins: Database result sets are perfectly tabular. TOON's schema mirrors SQL structure, making it natural for LLMs to process.

Analytics & Event Logs

Time-series data, user events, application logs sent to LLMs for anomaly detection or insights.

Typical Event Data

• Pageview events (timestamp, user_id, page, duration)
• Error logs (level, message, stack_trace, timestamp)
• Purchase events (user_id, product_id, amount, timestamp)

Why TOON wins: Event logs can be massive (thousands of rows). TOON's 60% token reduction means fitting 2.5x more events in the same context window.

LLM Fine-Tuning Datasets

Preparing training data for fine-tuning GPT, Claude, or open-source models.

Cost Impact

10K training examples (JSON)

~50M tokens

$400 training cost

10K training examples (TOON)

~20M tokens

$160 training cost

Why TOON wins: Fine-tuning charges per token. Smaller format = dramatically lower training costs + faster iteration cycles.

LLM-to-LLM Communication

When AI agents send structured data to each other (e.g., MCP protocol).

Why TOON wins: AI agents don't care about human readability. They only need parseable structure. TOON's efficiency improves agent-to-agent bandwidth.

When NOT to Use TOON

Traditional Web APIs

APIs between services, frontend-to-backend communication.

Why JSON wins: JSON has universal tooling support. Browsers have native JSON.parse(). Don't add complexity where it's not needed.

Deeply Nested Hierarchical Data

Config files, complex object graphs, tree structures.

Bad Fit for TOON

{
  "user": {
    "profile": {
      "settings": {
        "notifications": {...},
        "privacy": {...}
      }
    }
  }
}

Why JSON wins: TOON is optimized for flat, tabular data. JSON's hierarchical structure naturally represents nested objects.

Single Object / Small Arrays

User profile, single database record, config with 2-3 items.

Why JSON wins: TOON's efficiency comes from schema reuse. With only 1-3 rows, there's no benefit. JSON is simpler for small data.

Inconsistent Schemas

Array where each object has different fields.

Bad Fit for TOON

[
  {"type": "user", "name": "Alice", "email": "..."},
  {"type": "product", "title": "Laptop", "price": 999},
  {"type": "order", "orderId": 123, "total": 54.99}
]

Why JSON wins: TOON requires consistent fields across all rows. Varying schemas break TOON's efficiency model.

Quick Decision Matrix

Scenario	Format	Why
RAG with 100+ database rows	TOON	Massive token savings
REST API response	JSON	Universal tooling
Training dataset (10K examples)	TOON	Lower training costs
Nested config file	JSON	Better for hierarchies
Analytics events (1000+ rows)	TOON	Fit more in context
Single user profile	JSON	No benefit for 1 object
AI agent communication	TOON	Efficiency matters

Real-World Impact

Case Study: E-Commerce RAG System

Challenge: Product search chatbot needed to inject 200 product records per query

JSON Approach: 8,400 tokens/query × 50K queries/day = $12,600/day in API costs

TOON Switch: 3,600 tokens/query × 50K queries/day = $5,400/day

Result: $7,200/day saved ($216K/month) with zero quality loss

Best Use Cases for TOON Format