Loading…
You and AI, unified
Walk the Way
Substance over volume.
Proof over performance.
Human and machine, side by side.
Slow is not the enemy//Slop is
We use cookies to understand how you use this site and improve your experience. Privacy Policy
YounndAI Object Notation™
Data, intent, provenance, and thought — in one stream.
One package — the parser, generator, runner, and the rest of the toolchain.
npm install @younndai/yonVerifiable
Transport formats move data between machines. Prose and markdown are readable but unstructured. YON fills the gap between them — structured and parseable, line-oriented, readable by people and parseable by machines — and it holds what neither was built for: data, intent, provenance, and thought, in one stream.
Every record line is independently parseable. No global schema required. Enter mid-stream and understand. It is not a JSON replacement — different category, different job: YON is a format for writing meaning, not just moving values.
Built for a reader that can’t scroll back.
Every format before YON assumes random access. The new reader is a streaming model that can’t backtrack — so the data is there the moment the line arrives.
What is — the structured facts.
What it will do — the constraints.
What happened — the audit trail.
What it thinks — the reasoning.
Skills, protocols, and configs a model acts on directly — like the Open Skills YON ships.
A model writes YON and each line is usable the moment it lands — no waiting for a closing brace.
The same file a strict parser and a language model can use — no translation layer in between.
~5%
structural overhead vs minified JSON
~30%
leaner than prose, at scale
~99%
recovery under truncation
+22pp
budget-tier comprehension
Each number against its baseline. The comprehension gain is cheap-model economics — budget-tier models reading structure they couldn’t parse from prose — not a universal claim.
We run our own stack on it — and others we’re not naming yet.
The direction: authoritative data published agent-ready — structured and attributable — read without a translation layer.
We run our own stack on YON today — and we're building toward more of it speaking one structured stream.
Where it's heading: data, intent, provenance, and thought as one stream a system reasons in — first line to last.
Say less, get understood better.
YON and YounndAI Object Notation™ are trademarks of MARLINK TRADING SRL.
