They said weights are frozen. We made them grow with you.
Continual learning happens on your device: the model grows with use, remembers across sessions, and your data never leaves.
“We don't sell tokens. We set them free.”
How it works
Memory, learning, ownership — the brain of embodied AI lives in the device, not in someone else's cloud.
The devices around you generate data continuously — some from what you do, some from what they sense.
The device locally understands your patterns, forming a cognition of who you are.
That cognition is imprinted into the model's internal state — no weight edits, no fine-tuning, no cloud.
Your data physically never leaves your device.
Why it's different
Not fine-tuning. Not RAG. A recoverable learning state that travels with the model — on whatever runs it.
Learning is captured as a compact inference state — never a gradient step.
It survives sessions and restarts, is gated before activation, and rolls back in one step.
It lives at the inference-state level, so it isn't tied to a model family or a processor.
Validated end to end on consumer hardware.
Evidence and boundariesCross-silicon
The same learning and inference algorithms; only the execution backend changes. On Apple devices, learning closes the loop on the phone itself. On Intel, our runtime has shipped to a world-leading PC manufacturer.
iPhone / iPad / Mac
Core Ultra · CPU + Arc iGPU
More silicon platforms on the roadmap
No dependency on chip vendors' high-level APIs — we go down to kernels and compute graphs when needed.
Intel figures compare runs on the same machine with the same model, an Intel Core Ultra 5 225H. RTF = generation time ÷ audio duration; speech synthesis figures are medians from a 30-minute resident run.
Products
Four free apps are its proving ground — private, no accounts, no ads.
Research
We're looking for researchers and engineers who want to push the boundaries of on-device AI.