Research
Notes from the build.
Points of view grounded in what we actually ship — AI privacy tooling, local-LLM infrastructure, and cross-platform product work. We open-source the tooling layer and plan to publish our applied research.
Redaction you can trust: what we learned building Censory
Covering text with a black box is not redaction — the words are still in the file. Here is what actually removing them takes.
Making local LLMs actually usable
Running a model on your own hardware is the easy part; the friction is fitting it to the GPU, serving it cleanly, and getting agents to coordinate instead of idle.
Shipping across platforms from one core
Eve, Fit and Censory are very different products that share one delivery pattern: a web core, a mobile-first surface, and PWA install where a native shell would be premature.
Why we're open-sourcing our tooling — and plan to publish
Skynet Cluster and TurboProvider are already open source. That is a deliberate stance on where our infrastructure work belongs, and where applied research goes next.