Ternlight Packs a 5MB AI Model for On-Device Website Search
The package makes on-device AI search practical for any website, removing server costs and privacy concerns.
Reporting from 1 source: GIGAZINE.
Ternlight is a new package that lets developers embed a 5-7MB AI model into websites. The model runs on the visitor's CPU, enabling semantic search without external servers. Two variants are available: base (7MB) and mini (5.5MB). Developer Wen Shu Tang released the package on GitHub and npm, and is seeking contributors.
The demo site loads the model on access and performs searches with a few milliseconds delay on a laptop, according to the package's developer. The model is a distilled version of the small embedding model all-MiniLM-L6-v2, packaged in two sizes: @ternlight/base at 7MB and @ternlight/mini at 5.5MB.
Web developers can use ternlight to add semantic search or other embedding-based features to their sites. The computation happens entirely on the visitor's device, meaning no data leaves the browser. The package is available on GitHub and npm under an open license. Wen Shu Tang, the developer, is seeking contributors to improve performance and quality.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.