NeuroNAS is built on a "four chips, three zones" layered architecture. Capture, Transport, Decision and Application each stand alone yet work in seamless concert, closing the loop from EEG signal capture to whole-home intent control. The entire pipeline runs without ever touching the cloud.
intent → device
theta/beta neurofeedback
acoustic closed-loop control
SDK ecosystem
cognition
Lightweight Transformer + SNN hybrid inference (≤100M params/model)
Federated learning framework (local iteration + anonymous aggregation)
Neuromorphic chips: AKD1500 / Innatera Pulsar
Home neural-data private cloud (encrypted local storage + hot/cold tiering)
Home-grade on-device AI inference (NPU / Rockchip edge silicon)
Family member profiles + cross-generational neural-data linkage
accessory connectivity
home device connectivity
8–16 channels · graphene dry electrodes
4–8 channels · behind-the-ear dry electrodes
2–4 channels · head EEG
EEG data runs every AI inference locally on the NAS, keeping privacy 100% under your control. On-device processing cuts latency by over 60% versus the cloud.
The BrainChip AKD1500 uses an event-driven architecture, firing computation only on "spike" activity, so standby power is near zero and delivers 24/7 continuous sensing. The Innatera Pulsar cuts power by roughly 500×.
No cloud required: thought-as-command runs through a local Matter gateway to control whole-home devices directly, and works offline. A Thread border router plus Zigbee bridging delivers seamless cross-brand orchestration.
The "four chips, three zones" mainboard, the full chip list, channel counts and protocol stack, all out in the open.
View hardware specs →