Ternlight

Ternlight is a compact embedding model designed to run directly in the browser using WebAssembly (WASM), with a total size of only 7 MB. This allows developers to perform semantic search, clustering, or similarity computations on the client side without sending data to a server. The model is optimized for low latency and privacy, making it suitable for applications where data must stay local. Based on community signals so far, Ternlight appears to be a fresh launch, with a demo available at ternlight-demo.vercel.app. The lightweight footprint enables integration into web apps, browser extensions, or edge functions. While specific performance benchmarks and API details are still emerging, the core value proposition is clear: a self-contained embedding model that eliminates server dependencies for vector generation.

Ternlight appeared on Hacker News with a demo link, highlighting its tiny 7 MB size and browser-based WASM execution, signaling a fresh launch in the embedding model space.