Expand description
§aura-toon
Lightweight encoder for TOON — Token-Oriented Object Notation.
TOON is a JSON-compatible serialization format designed to reduce token count when feeding structured data to LLMs. For typical payloads it produces 30–60% fewer tokens than equivalent JSON, with a tabular form for arrays of homogeneous objects.
Spec: https://github.com/toon-format/spec
§Example
use serde_json::json;
let value = json!({
"snapshots": [
{"file": "main.rs", "trigger": "watcher", "ts": 123},
{"file": "lib.rs", "trigger": "mcp", "ts": 456}
]
});
let toon = aura_toon::encode(&value);
assert!(toon.contains("snapshots[2]{file,trigger,ts}:"));§Caveman Mode
For natural-language strings, caveman() strips articles, filler words,
hedging, and pleasantries — producing terse, fragment-style output that
uses 20–40% fewer tokens while preserving all technical meaning.
let verbose = "I would be happy to help you with that. The function is currently \
failing because the variable is not properly initialized.";
let terse = aura_toon::caveman(verbose);
assert!(terse.contains("function"));
assert!(!terse.contains("would be happy"));Originally extracted from Aura, the semantic version control engine.