Getting started

Fingerprint text. Compare similarity in one call.

Glyph builds MinHash signatures from text. You use Create() to fingerprint, then Compare() to score overlap.

Install

npm install @koda.oss/glyph

Node.js 18 or newer is required.

First compare

import { Create, Compare } from "@koda.oss/glyph";

const a = Create("the quick brown fox jumps over the lazy dog");
const b = Create("the quick brown fox leaped over the lazy dog");

const { similarity, distance, matches, size } = Compare(a, b);

Field	Range / type	Meaning
similarity	0–1	Estimated Jaccard similarity
matches	integer	Matching signature slots
distance	integer	size - matches
size	integer	Signature length

Mental model

1. Tokenize input text into words.
2. Build a feature bag: tokens, unigrams, vgrams.
3. Run bag MinHash to produce a Uint32Array (glyph).
4. Compare glyphs by counting equal slots.

Identical normalized text gives similarity: 1. Unrelated text is near 0.

What Create returns

{
  "version": 1,
  "glyph": "<Uint32Array>",
  "createdAt": 1710000000000
}

• version — create pipeline version.
• glyph — the fingerprint bytes.
• createdAt — milliseconds since epoch (changes each call).

You can pass a full record or record.glyph into Compare and Serialize.