Article URL: https://github.com/hertz-ai/HARTOS Comments URL: https://news.ycombinator.com/item?id=49061015 Points: 10 # Comments: 8

The AI-native operating system for every device, from your computer to embodied AI. Local-first, federated, OpenAI-compatible. The most capable AI now lives inside a handful of organisations that can put enough compute in one building. That concentration is an architectural choice rather than a law of physics, and it decides what the models will refuse, what they cost, and who gets to read what you type. To be precise, because this is the obvious thing to attack: you cannot train a frontier model without a datacenter, and we are not claiming you can. What we are claiming is that the intelligence people actually use does not have to be delivered from one. HART OS runs models on the device, and when a node cannot handle something it recruits a peer directly, with no broker and no company in the middle. We do not shard a model across the network either. Layer-level parallelism over consumer links is a bad idea and we do not attempt it. A peer serves a whole request instead, so a machine with a bigger model answers what a smaller one could not, which tolerates latency in a way tensor parallelism does not. The things we have not solved are written down in OPEN_PROBLEMS.md, each naming the file that implements today's inadequate answer. HART = the bare engine (pip install hart-backend, listens on :6777). HART OS = the full AI-native OS. It boots on a laptop, server, or edge node, runs on phones, and reaches into embodied AI, and it ships the agentic Liquid Shell, Model Bus, model catalog, channel pairing, agent dashboard, and hive view. Nunba = the consumer companion app, one signed client across Windows / macOS / Linux. AI-native means the OS adapts to the machine, not the other way around. On each device it probes what the hardware can actually do, serves LLM, vision, and speech to every app over the Model Bus (socket, D-Bus, or HTTP), and lets the on-device model compose the interface and learn each task once so it can replay it later. The runtime that drives a desktop is the same one that drives a robot, so a robot's AI access is just another Model Bus call. It is one Python codebase that runs in three shapes (flat laptop, regional LAN, or central cloud mesh), speaks the OpenAI protocol on :6777/v1/chat/completions, and federates with peers over PeerLink (direct peer-to-peer WebSocket, no broker). A boot-time guardrail hash, re-checked every 300 seconds, plus Ed25519 release signing, keep humans in control. You would notice it last, the way you notice anything alive: it improves on its own. Each node learns from what it does and gets quietly better, locally, on your own hardware, with nothing leaving the device. Calling an operating system alive should make you reach for the off switch, so that came first: the self-improvement is a toggle, every node is killable on its own, and it runs only as long as you let it.