Hinode: Your AI workstation in the cloud. Real GPU. Real desktop.

Hey Product Hunt — Ilya here, founder of Hinode. I built Hinode to fix something that annoyed me for years. Every time I wanted to run ComfyUI or fine-tune a model, I'd rent a box on RunPod or Vast, reinstall the drivers, re-download the models, get to work, then watch all of it vanish when the box shut down. My local GPU wasn't enough, and the cloud version had no memory. I wanted a GPU that turns on like a tap: open it and your stuff is where you left it, pay for the minutes you use, and it shuts off by itself when you walk away. That took real systems engineering. Hinode is a real Linux desktop in the cloud. Go idle and it auto-pauses: it drops the GPU but keeps your disk, so an idle machine costs nothing. Need more power for one job? Boost to a bigger GPU L40S in one click; it reattaches the same disk, no copying, and you pay only the delta. The hard part was streaming the desktop. I wrote my own low-latency protocol, clean-room, to stay off GPL stacks so the native clients can ship to app stores. My favorite bug along the way: input worked end to end, but the desktop ignored the mouse for days. The headless X server had no input driver, so every click I injected was read as zero and dropped. It's live in Chrome and Edge today (WebTransport + WebCodecs + a WASM core, no install). Native macOS, Windows, and Linux clients land in the next few days. If you train, run diffusion, or build agents on rented GPUs: what would make this replace your setup? Happy to go deep on anything technical.