Concept — YON

# YON

## Definition

A constructed notation for AI cognition that separates **grammar** from **vocabulary**, enabling domain-pluggable LLM interoperability. Unlike free-form prompts or rigid JSON schemas, YON is stream-first — designed for the way language models actually generate text (token-by-token, forward-only) — while staying parseable, composable, and machine-readable.

## Coined by

Alexandru Mares (allemaar)

## First published

2025 (specification work in progress through 2026).

## Canonical artifact

- Project hub: [[yon.younndai.com|YON project]] (`alm-os/Projects/YON - yon.younndai.com/`)
- Domain: [yon.younndai.com](https://yon.younndai.com)
- Spec, parser, generator, converter, runner — see project hub
- License: Apache 2.0 (open source)

## Related concepts

- [[younndai|YounndAI]] — sibling work
- [[shin|SHIN]] — the YON compiler
- [[zen|ZEN]] — the YON runtime
- [[ken|KEN]] — the YON witness
- Stream-first notation
- AI cognition / agent notation
- [[yon|Cluster: YON]] — parent Cluster

## Why it matters

The block-format problem (essays, JSON, XML) requires a model to produce a whole document correctly before any of it can be used. YON is line-independent and stream-first — each line is a self-contained record that can be parsed, validated, or acted upon as it arrives. This matches how LLMs actually produce text and removes a class of failure modes (one bad token corrupts the whole output) that block formats inherit by design.

Grammar / vocabulary separation is YON's other distinctive move: the grammar is fixed and minimal (the structural language), while vocabularies plug in per-domain (`yon.sai`, `yon.content`, `yon.tasks`, ...). This lets one notation serve many domains without forking the parser.

## Status

`active` — under specification; parser implemented; generator and runner in development.