Announcing Project Glass 2.0
ProjectGlass 2.0 is a search engine I've been building, and it comes down to three things I care about more than anything else: proving it's actually getting better instead of just saying so, staying out of your business as much as I possibly can, and working for whoever shows up, no matter what they're browsing on.
Proving it, not promising it
Every search engine that claims it's "getting smarter over time" is basically asking you to trust it, no real way to check. So instead of one ranking system and a promise, ProjectGlass runs three at once, side by side, and every search gets anonymously dropped into one of them.
Arm A is the one actively being tinkered with, tested against real traffic. Arm B is whatever previously beat everything else, the reigning champion. When A beats B by a real margin, not just noise, it becomes the new B.
Arm C is the piece I care about most. It's frozen, locked exactly as it was the day this launched, forever. If you only compare new versions against whatever's currently winning, you can drift a long way from where you started without ever knowing if all those changes added up to anything. Arm C keeps that honest, a permanent check against doing nothing at all.
None of this needs to know who you are. A session gets randomly dropped into A, B, or C, stored in a cookie that just says which letter, nothing more, checkable in your dev tools right now. What gets logged is "a click happened, on this result, in this position, in this arm." Not who did it. The moment your session ends, that's gone too.
And there's no AI writing anything here. Machine learning drives ranking, some NLP parses what you're asking, but nothing generates an answer for you. It ranks. It doesn't talk at you.
Accounts, but no email, ever
You don't need an account to search, no login, no cookie beyond the arm letter. If you want one, it stays small on purpose: pin yourself to an arm, mute or boost certain sites, get a light personalization layer built off your own clicks, kept separate from the experiment above.
What it will never have is an email field, which means no password reset either. Lose your password, you're making a new account. Harsh, I know, but intentional. A reset needs somewhere to send a link, and once there's an email tied to an account, there's a permanent thread connecting a real inbox to everything that account's ever done. I'd rather that link never exist than promise to protect it forever.
It just has to work
None of the core stuff, searching, browsing results, switching themes, needs JavaScript at all. The only thing that does is logging in, and even that's optional. I actually tested this in a text-only browser from the early 90s and fixed whatever broke.
Where this actually stands
I'm not going to oversell it. The three-arm setup only means something once there's real traffic behind it, and there isn't much yet, so what it's telling me today is closer to a guess than a fact. I trust the design, I just don't have months of data yet, and I won't pretend otherwise.
If this sounds interesting, go try it and tell me where it breaks, that's what arm A actually needs. Once there's enough real data to talk about, I'll write the longer, more technical version, the architecture, the ranking internals, all of it.
You can access ProjectGlass at https://projectglass.xyz