HART OS - Open-source AI-native OS for local-first agentic runtime
Show HN: HART OS – an open-source AI OS built so frontier AI needs no datacenter
HART OS is an open-source, AI-native operating system designed to run frontier AI models directly on local devices, eliminating the need for datacenters. It transforms inference into a core system service accessible via an OpenAI-compatible API, enabling seamless operation across laptops, servers, and embodied AI robots. Featuring a federated peer-to-peer architecture and a self-improving auto-evolve loop, HART OS allows applications to dynamically access local LLMs, vision, and speech capabilities without managing API keys or model downloads. This local-first approach ensures privacy while providing a unified runtime that adapts to hardware constraints and learns from user interactions.
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, and lets the on-device model compose the interface and learn each task once so it can replay it later.
- tpoacher
I think what the author is trying to say is that the turboencabulator's vertex manifold buckler dimensional engine leverages bifrontal synergies spanning several plasmic thresholds.
This enables an Agentic Solution Plexus to surplant inter-connectomic data access completely locally over the internet, securely, in a federated sandboxed reinforcement learning evolutionary cow-milk optimisation harness, complete with Lorentz-Newton manifold compatible titanium bearings.
Sounds exciting!
- DougN7
You will get MUCH more uptake and discussion if _you_ write that main github readme page. Many, and I’m becoming one too, won’t bother reading if you couldn’t be bothered to write.
- KaiserPro
This seems a little, messy?
is the readme aimed at humans? because its a bit scatter gun.
Is it a model orchestrator? or a wrapper around docker/nix/cgroups+chroot?
Like what makes it an OS?
You rightly point out that shared memory across wifi/crappy ethernet is a bad idea, but then how do you get around that?