About Vireon

A language system built from connected Cells and visible evidence

Vireon is designed to learn relationships through sentences, preserve those relationships inside SentenceCells and later extend them through structured semantic fields.

Updated: 29 July 2026

The present architecture

Vireon does not store every learned statement as a duplicated block inside each word. Instead, reusable WordCells and ActionCells retain stable identities, while SentenceCells preserve the original sentence, token identities, grammatical roles, contextual resolution and the relationship between those Cells.

This makes a SentenceCell both a memory of the source and evidence for why an answer is valid. A word such as brother can be reused by many sentences without carrying backlinks to every sentence in which it appeared.

Source text is recognised and divided into source-ordered sentences.
WordCells and ActionCells provide reusable identities.
SentenceCells store compact sentence-local role connections and contextual meaning.
Re-reading reconstructs temporary discourse state from the original source order.
Answers are returned from matching SentenceCell evidence.

Current stable achievement

The V1.0 baseline demonstrates deterministic re-reading and connected retrieval across multiple SentenceCells. Question order may be canonical, reversed or deterministically shuffled while the reading order remains the original text order.

V1.0 is therefore the stable foundation for the next phase: a much larger Cell population and sentence-driven semantics.

The semantic direction

The nodes are the Cells themselves. The planned semantic levels are organised fields within or attached to each Cell:

key → class → subcategory → category → subdomain → domain

Learning still begins with sentences. Sentence usage shows which meaning is active and which Cells are related. The category fields do not replace SentenceCell evidence; they create additional routes through which related Cells can later activate one another.

What Vireon does not yet claim: general semantic reasoning and natural summary generation remain future work. Those functions will return only after the vocabulary, category fields and definition-based meaning network are sufficiently developed.