Today we are announcing the biggest shift in CORe's history. We are moving away from creating cloud AI models, Genesis and the lineup that came before it, and toward something we believe matters more: making local AI easy. The next CORe platform is fully local-first, and it will be entirely open-source.
This decision came from two places. The first was an experiment that genuinely scared us. The second is a conviction we have held quietly for a while: the cloud AI bubble needs someone pushing against it, and we intend to push.
The experiment that changed our minds
A few weeks ago we ran an internal experiment. With the explicit consent of the employees involved, we built a small internal tool that assembled a digital profile of a person purely from their chats and stored memories. That is the same raw material every cloud AI provider sits on.
The results scared us. For one employee who had barely ever used the platform, the profile still pinned down the state they live in, the teams they root for, and what they do for a living, reconstructed from a handful of casual conversations. We showed the employees their profiles, and then we deleted the tool and everything it produced the same day.
The lesson was impossible to unsee: anything you tell a cloud AI is raw material for a profile of you, assembled not by malice but by default, because the data is simply there. The only architecture that truly respects you is one where your data never leaves your machine.
Pushing against the cloud AI bubble
The experiment alone was not the whole story. It landed on top of a frustration that had been building for years. Cloud AI has become a bubble of ever-larger models behind ever-pricier subscriptions, where every prompt you write flows through someone else's datacenter, on someone else's terms, metered by the token. The entire model of the industry assumes you will rent intelligence forever and hand over your words as the cost of admission.
That assumption is breaking. Open-weight models now run beautifully on ordinary consumer hardware, and the gap between "local" and "cloud" shrinks every month. The bubble survives on the story that you have no alternative. We are building the alternative, in the open, and we are happy to be the company that demonstrates it.
What local-first means
The next CORe runs on your hardware, not ours:
- Your chats, memories, and settings stay on your machine. Nothing is streamed to a third party unless you explicitly point the platform at an outside provider.
- The models run locally through whatever runtime you prefer. The platform is the interface; your computer is the datacenter.
- It works offline. No account wall, no usage metering, no API bill quietly growing in the background.
- It's entirely open-source, so "your data stays local" is a claim you can verify, not one you have to trust.
The platform
It is the CORe you already know: the same interface, the same memory system, the same attention to design, rebuilt to run locally. API configuration moves into the admin page, so pointing the platform at a local runtime takes seconds. A built-in local AI marketplace helps you discover and set up runtimes and models, whether this is your first local model or your fiftieth.
What happens to Genesis and the cloud models
We are winding down new cloud model development. The existing lineup stays available while we transition, and we will give plenty of notice before anything changes. Our focus, and our future, is local.
CORe started as a bet that AI could be aligned with its users. Going local-first is the purest form of that bet: the model works for you, on your machine, and answers to no one else.