How we got here
The tools kept getting smarter.
The rules kept getting harder.
Sorvian started from a question we kept hearing from security, engineering, and legal teams at the same company: “Can we actually use this AI without ending up on the front page?”
The answer was usually some version of “not really.” Teams either sent everything to a public model and crossed their fingers, or redacted so much that the model came back with confident, useless answers.
We thought the tradeoff was avoidable. A small group of us started sketching what a middleware layer would look like — something that could understand a company's data as deeply as a frontier model, but without the data ever leaving the company's walls. That sketch became the distillation pipeline.
A few years later, that's still what we're building. A bit bigger, a lot more polished, and still pointed at the same problem: keeping the answers without giving up the data.
The team
Builders, researchers, operators.
We wanted to work on the problem every AI team is going to run into — and we wanted to be the answer when they did.