GCFGraph Compact Format

The most token-efficient wire format for LLMs. 79% fewer input tokens than JSON. 75% fewer output tokens. 100% comprehension accuracy at scale.

Get Started

Try the Playground

GCF vs TOON

79% fewer input tokens

Positional fields, local IDs, and hierarchical grouping eliminate per-record overhead. At 500 symbols, JSON uses 53,341 tokens. GCF uses 11,090.

75% fewer output tokens

LLMs produce valid GCF with a 3-line primer. 52% smaller output than TOON. Proven at 5 to 100 symbols with 100% decoder validity.

Optimized for the reader that matters

At 500 symbols, JSON (the "readable" format) drops to 66.7% accuracy. The LLM miscounts records, drowning in structural noise. GCF scores 100%. Human-readability and LLM-readability diverge at scale. GCF optimized for the one that matters.

Gets cheaper every call

Session deduplication (92.7% savings by 5th call) and delta encoding (81.2% on re-queries). No other format has these. They compound.

Beats TOON on TOON's benchmark

34% fewer tokens on mixed-structure data, 44% on semi-uniform, 3% on flat tabular. Tested with their datasets, their tokenizer, their library.

Three languages, zero dependencies

Go, TypeScript, and Python. Graph profile (encode) and tabular profile (encodeGeneric). CLI included. MCP proxy for zero-code adoption.