May 7, 2026
We're offline in a good way. Li/o is now self hosted. Local. Version Six Point Oh damn. It's fine-tuned on even better data. Scarily accurate (F1: 89%). And now we can take it into a cave or a focus group, whichever comes first. It's been an ordeal. Beyond the data research and fine tuning, we built the whole thing, web apps and all, from scratch. No APIs phoning home. No cloud with its hand out. Just a lean, mean local model.
The benefit? Our hard-earned training data stays ours. Client data stays private. For a model built around the sanctity of real, unfiltered human language, that matters more than we can overstate.
When we started building Li/o, we were charting something completely new. And like most explorers who don't have a map, we built the thing we could build, not necessarily the thing we ultimately needed. We leaned on hosted infrastructure. We made API calls. We stood up databases. Dependencies stacked up like rush hour on I-94. It worked. But it came with baggage, and frankly, some explaining to do every time client data left the building.
So we fixed it.
We know what you're thinking. "That sounds hard to scale as a SaaS."
You're not wrong. But just like you wouldn't hand someone a DSM and call them a psychologist, Li/o is designed to work alongside people who understand cognitive psychology. Who can read what the model surfaces and know what to do with it. The people were always the most important part of this. We just made sure the technology finally matches that belief.
Ding dong Li/o's dead. Long live Li/o.