Our story

The assistant a doctor
is allowed to have.

I watched coding agents change what one person can do. Then I noticed who they couldn't reach. My wife is a doctor. This is why I'm building the whole stack — and why the story is still being written.

Five jobs, one person.

I do development and design, and I also answer for the company's finances, payroll, HR, and sometimes its legal. One person, five hats. AI has carried roughly 90% of that work for a long time. Coding agents — Claude Code, Codex, OpenCode — changed what one person can do. I wanted that for the person next to me.

The people with the most sensitive data have the least access to AI.

My wife is a doctor. Patient data cannot leave the environment where she works — not to a cloud, not to a chatbot, not to someone else's servers. So the entire category of AI built on "send it to us" simply doesn't exist for her. That's most of AI. The profession that guards the most important data on earth is the one the industry can't serve.

Most of the work doesn't need a frontier model. It needs a harness.

Building my own assistant taught me something I wouldn't have believed: the model is rarely the bottleneck. What matters is the harness around it — memory that retrieves without burning tokens, data that is structured instead of pasted, accounts that become tools, a file that says who the agent is. With the right harness, a small local model — running on the machine you own, nothing leaving it — can handle the majority of the work. That's the bet this whole stack is built on.

Every layer of the harness, open source.

Each layer runs local. Each layer is inspectable. The model is your choice — that's the most personal decision an assistant makes, and it should be yours.

I can tell you why I'm building it.
I can't tell you she's using it yet.

My wife hasn't used it yet. The story doesn't have its ending — that's the part I'm still building, and I won't pretend otherwise. But the test is the point: if an assistant works for a doctor — someone who can't afford for it to be wrong, and can't afford for it to be shared — it works for anyone whose data isn't theirs to give away. It will fail, or succeed, in public.

Medicine · Law · Finance · HR

If your data can't leave your machine, this is being built for you.

Tell us your constraint. It's our roadmap.