Ternlight – 7 MB embedding model that runs in browser (WASM)

A new project called Ternlight just strutted onto the internet with a bold promise: fast text search in your browser, no server, no account, no sneaky data trip to the cloud. The big flex is size — just 7 MB, with a 5 MB mini version — which had people doing a double take because that means a website can understand and sort text locally on your own machine. In normal-human terms: it helps apps find related text without sending your words off to somebody else’s computer.

But the real action was in the reactions. Creator soycaporal showed up with the classic humble-brag energy: this was a hobby project, but also one where they built the whole engine themselves and squeezed it into the browser. That instantly won admiration — and skepticism. One of the strongest practical takes came from aetherspawn, who basically asked, “Cool, but can we skip the slow setup and do that part ahead of time?” Translation: people love the speed, but they’re already poking at the startup pain.

Then came the comic relief. dirteater_ complained the demo unexpectedly made their computer fans "go crazy," which is exactly the kind of accidental chaos the internet lives for. Meanwhile, esafak pushed a bigger hot take: instead of every project inventing its own workaround, why not a standard browser feature for this stuff? And in true comment-section fashion, another user immediately rolled in with a live example, saying they already plugged it into an offline search tool. So yes: part applause, part feature requests, part browser-policy debate — and a little bit of laptop jump-scare comedy.