Neywa Labs is coming out of stealth with infrastructure for building and deploying AI that runs on Apple Silicon. Cloud requests still leave the device, depend on a provider staying available, and stop when a usage limit or a bill gets in the way. The company is aimed at developers and product teams who want models, agents, and app features to run on Macs people already own, with data staying local and no API key required. The base layer is an open-source Engine. MLX-VLM handles inference and fine-tuning of vision-language and multimodal models on Macs, and MLX-Audio covers speech, transcription, and generation. The company says those libraries support more than 5,000 models. Nativ, the Mac app built on that work, points coding agents such as Claude Code and Cursor at a local model, and adds offline voice, meeting notes, image and document understanding, video understanding, and image generation. Machines can also be clustered so their memory is pooled for models that do not fit on one device. Prince Canuma, co-founder and CEO, created MLX-VLM, MLX-Audio, and Nativ, and is building the company with Sebastien Michaud, co-founder and COO, and Lucas Newman, co-founder and CTO. The launch turns that open-source work into a path for teams to put on-device AI into their own products, including models from labs such as Google DeepMind, Cohere, and Liquid AI.