Ternlight: 7 MB embedding model runs entirely in the browser

A new embedding model compressed to 7 MB runs via WebAssembly in the browser, eliminating the need for server-side inference. For indie developers building search or similarity features, this removes both infrastructure costs and latency—the entire model ships with your app.