News
We do not publish often. When we do, it is here. For research write-ups, see the research page.
We are expanding our distributed inference fabric across additional regions in Europe and North America to meet growing demand for real-time conversational workloads. The platform now serves millions of conversations daily across a hand-tuned mix of NVIDIA and AMD accelerators.
Our in-house orchestration layer for model routing, token budgeting, and graceful degradation is now serving the majority of production traffic. It replaces an off-the-shelf serving stack that could not meet the latency requirements of multimodal conversation.
Voice synthesis and real-time image generation joined text in the conversational stack, with sub-second time-to-first-audio and reusable latent caches across a session.
Persistent memory for conversational continuity ships to all users. Companions remember the people they speak with, not just the last message.
Nastia begins as a solo project. One founder, one model, one product, built from a conviction that the consumer AI space had room for something gentler than the assistants on offer. The team has stayed deliberately small ever since. We work closely on every layer of the stack and ship product directly.
