Playground¶

Try AGON encodings in real-time. Enter JSON data and see how each format encodes it, with live token comparison.

JSON Input

JSON

Token Comparison

How It Works¶

The playground encodes your JSON data using all four AGON formats simultaneously:

Format	Best For	Typical Savings
JSON	Baseline / irregular data	0% (baseline)
AGONRows	Uniform arrays with 2-10 fields	30-60%
AGONColumns	Very large objects (100+ fields)	20-50%
AGONStruct	Repeated nested patterns	10-30%

The Auto Selection indicator shows which format AGON would choose based on:

Minimum 10% savings threshold vs compact JSON
Data structure analysis (uniform arrays, wide tables, repeated patterns)

Format Selection Notes¶

Rows wins for most cases: - Uniform arrays with consistent fields (even with 10-20 fields per record) - Example: 788 employee records with 12 fields each in benchmarks

Columns wins for very large objects: - Single objects with 100+ fields (like quote.json with 170 fields in benchmarks) - Type clustering benefits outweigh format overhead at this scale

Struct wins for repeated patterns: - Objects with the same structure appearing 3+ times - Common pattern: {fmt, raw} for formatted/raw value pairs

Try It¶

Token counts use js-tiktoken with o200k_base encoding (GPT-4o default)
The JavaScript encoders match the Python/Rust AGON library behavior
Token counting matches Python's tiktoken library exactly