We have our first paying customer at Hinode. He almost didn't sign up.
He builds open-source AI video tools and ships them on Hugging Face. He prototypes on a maxed-out Apple-silicon laptop. But a laptop can't touch a real CUDA path, so to actually run his code he ships it to a live Hugging Face Space and debugs it there. Every fix means a redeploy, a full model preload, and a scroll through the logs. He was using redeploys as a debugger, and in his words the loop was "tiring and time-consuming."
That loop is what Hinode is built to remove. It's a persistent Linux workstation with a dedicated NVIDIA GPU, in your browser or a native Mac or Windows app. Run the script, read the stack trace a few seconds later. Your environment and weights stay put between sessions instead of preloading from scratch.
His video models need 96 GB of VRAM. Our top tier gave 48. He told me so plainly, which is exactly the feedback I want. So we built the tier he needed. The next day, Hinode had Power: a dedicated RTX Pro 6000 Blackwell with 96 GB. The models that wouldn't even load at 48 GB now load and run, on a machine that's yours instead of a Space you redeploy into.
His reaction: "Damn, that was fast. This is some serious witchcraft." He signed up the same day.
Our first sale taught me something simple: talk to one real user, then build what they actually ask for. He named the ceiling; we raised it.
