Your own AI,
built at home.

Every assistant I tried wanted the same thing before it would help: my life. My mail, my notes, my calendar. Hand it over and it gets useful. Keep it and it stays a stranger.

Here's the part that never sat right with me. All of that ends up on computers I've never seen, run by people I'll never meet. When they change the rules, I live with it. When they make a mistake, I pay. And the AI on my own Mac is already good enough that none of it had to leave.

So I asked the simple question. What if it just stayed home?

It started in August 2026, in a small room in Romania, under the working name Heirloom: a family of small, open models that learn from the person using them and run on that person's own computer. Not one giant model in the cloud. A few modest ones at home, each good at the everyday things. What did I miss. What did I promise. Draft the reply.

In September, Emergent Ventures backed it, for one reason: combining your life with AI is clearly the future, it has a serious privacy problem, and this is a way around it. That sentence became the test for every decision since.

In October, Heirloom became Sublunar. Sublunar means beneath the moon: the ordinary world, down here, where life actually happens. Your desk, your kitchen table, your Mac.

Nothing leaves unless you send it.

Glim reads the mail your Mac has already downloaded. It answers from a memory it keeps on your disk, which you can read, edit and delete. When it needs the web, it shows you the exact question with your names and numbers stripped out, and waits for your yes. One switch turns every tool off. That isn't a line in a policy. It's how it's built.

We build on open models and say so, every time. The engine is open. You pay once for the app and you own it, and if you ever stop paying you keep everything you have.

I use it every day at my own desk. You get the same Glim I do.