Glyph 1.0: Fast MinHash Text Fingerprinting
We are excited to announce the release of Glyph 1.0, a lightweight TypeScript toolkit for generating, indexing, and comparing content fingerprints using MinHash.
What is Glyph?
Glyph turns text into fixed-size signature hashes (MinHash), enabling rapid similarity comparisons, in-memory indexing and ranked querying, document chunk spotlighting, and text completions.
Key Features in 1.0
- Core Fingerprinting & Comparison: Generate MinHash glyphs with
Create()and compare Jaccard similarity in sub-milliseconds withCompare(). - In-Memory Query Indexing: Wrap indices with
query.New(idx)and perform ranked similarity searches across thousands of documents. - Collections & Aggregators: Aggregate labeled document glyphs into a single queryable probe using slot-wise MinHash operations.
- Spotlight Document Chunking: Chunk and rank relevant passages within long documents.
- Glyph-Guided Completions: Provide fast next-word Markov suggestions seeded by glyph signatures.
- Official NPM Scoped Package: Published as
@koda.oss/glyph.
Check out the Glyph Documentation and the Migration Guide to get started!