ByteNova

ByteNova, formerly ClustroAI, builds an edge AI framework that argues for moving model inference off the cloud and back onto user devices. Its containerised runtime schedules adaptively across GPUs, NPUs and CPUs, runs models locally so data never leaves the machine, and therefore cuts latency, bandwidth and cloud cost while suiting privacy-sensitive settings.

Two layers sit above it: a model layer with dialogue models trained for companionship, speech synthesis and recognition engines, and behavioural models that learn a user's routines and preferences; and an application layer in the desktop companion N.O.V.A, which keeps long-term conversational memory, develops a stable persona, allows custom appearance and voice, and connects to plugins and MCP tools for reading market data, monitoring social accounts or automating daily workflows. Developers can publish plugins and earn incentives by installs, and a user's identity and memory are bound to a soulbound credential.