Hinode: a cloud GPU computer that keeps your work
A Linux workspace in the cloud that keeps your files, apps and models. Add GPUs with up to 96 GB of VRAM for ComfyUI, training or batch jobs, and pay by the minute. Work over SSH, in VS Code or with your AI assistant.
Eligible new accounts get $10 to try after a $1 card check and refund.
Workspace $0.24/hr · Machines from $1.40/hr ·
Pauses itself when idle
Sign up here, then work from your computer.
ComfyUI · open in your browser
Job · train.py — 2h 11m
Job · label.py --part 3 — 1h 10m
The problem
Rented GPUs forget everything
A pod is built to be thrown away. Your setup is not. The same evening, two ways.
sent 31.4 GB · 22 min
$ python generate.py --batch 24
done · 24 images · 214 s
Terminating instance and cleaning up...
A pod is gone when you stop it. Models, custom nodes and outputs go with it. Tomorrow you upload 31 GB again.
loras/
outputs/
Hinode pauses instead. The card stops billing, the disk keeps everything, and Resume brings back the same files and apps.
How it fits together
One workspace. As many machines as the work needs.
Set up your workspace once. Every machine starts with that setup and reads the same shared folder.
The workspace
Your home folder and the apps you install. It pauses when you leave and starts again when you come back.
Machines
A whole card, 24 to 96 GB, beside the workspace. Each starts from your latest snapshot, so ComfyUI is already there.
The shared folder
Models, datasets and outputs, in one place every machine reads. Delete a machine and they stay.
Machines
Rent a GPU with 24, 48 or 96 GB of VRAM
Pick the VRAM and RAM the job needs. The machine starts with your apps on it and bills by the minute.
$ comfy run flux_video.json
loading wan2.2 · 31.2 GB
CUDA out of memory
job killed · 11 min lost
The job dies. The model needs more memory than the card has, and a bigger pod starts empty.
Attach a machine
GPU memory
System RAM
$ comfy run flux_video.json
loading wan2.2 · 31.2 GB
✓ loaded · 31.2 / 96 GB
rendering 81 frames ...
Start the card the job needs. From $1.40 to $7.41 an hour. Pause it or delete it when the render is done.
Apps
ComfyUI, vLLM and Ollama, installed in one click
Install from the workspace screen. What an app needs comes with it. Save a snapshot, and the next machine starts with all of it.
ComfyUI
Installing · 64%
PyTorch
~3 GB
JupyterLab
~50 MB
AI Toolkit
~8 GB
+ 6 more in the catalog
2 apps are not in a snapshot yet
- v4torch 2.5 and the new LoRA nodes
Shared folder
Write a checkpoint on one machine. Use it on the next.
Models, datasets and outputs live in one folder. The workspace and every machine read it at once, so there is nothing to copy and nothing lost when you delete a machine.
1train-48 finishes training and saves a LoRA
$ python train_lora.py --steps 2000
step 2000/2000 · loss 0.041
saved → ~/shared/loras/
portrait-v3.safetensors
2comfy-96 finds it in its LoRA list
lora_name
320 MB/s
reading one big model over four streams
440 MB/s
reading it again, from the machine’s cache
$0.10
per GB a month, only for what is in it
Local home
Python environments and git stay laptop-fast
Jobs
Start a LoRA training run and go to bed
One command gets a machine and runs your script on it. The log stays readable, the machine pauses itself when the work is done, and we email you the result.
11:42 pm · your laptop
~ — zsh$ hinode run --size 48x64 -d \
"python train.py"
job j-7f2 started
2:46 am · while you sleep
python train.pySucceededtrain-48 · 3h 04m · $9.72
train-48 paused itself at 3:01 am8:05 am · your inbox
HinodeYour job finished: python train.py
3h 04m on train-48, exit 0. The machine paused itself.
What you pay for
Pay by the minute. Idle GPUs pause themselves.
No plan and no allowance: you pay by the minute from credit. Idle computers pause themselves, disks grow on their own, and the bill names every machine.
September so far
$74.11
- studio · workspace · 64 h$15.36
- comfy-96 · 96 GB machine · 3 h 10 m$17.83
- train-48 · 48 GB machine · 3 h 04 m$9.72
- Disks · 96 GB written$14.40
- Shared folder · 156 GB$15.60
- Traffic out · 12 GB$1.20
bench-24 pauses in 2 minutes
Nobody has touched it for 13 minutes. After 15, it pauses and the hourly charge stops. The disk and everything on it stay.
Home folder on comfy-96
grew 128 → 256 GB · no restart
Ways in
Use it from a Mac, a PC, SSH or Claude
SSH and VS Code for the terminal, the Hinode app for your files and apps, and your AI assistant for everything else. One workspace behind all of them.
Claude · connected to Hinode
Train the portrait LoRA overnight on a 48 GB card, and pause it when it is done.
hinode_submit_job
train-48 attached · job j-7f2 running
Done. train-48 (48 GB VRAM, 64 GB RAM, $3.17/hr) is running python train_lora.py from your v4 snapshot. It pauses itself 15 minutes after the job ends, and you will get an email with the result.
Add Hinode in Claude's connector directory, https://mcp.hinode.sh anywhere else, or hinode mcp on your own computer.
Pricing
Cloud GPU pricing, by the minute
Eligible new accounts get $10 to try after a $1 card check and refund.
WORKSPACE
$0.24/hr
1/8 of an NVIDIA L4, your files and your apps. Pauses itself when you leave.
- Disk
- $0.15 / GB-month written
- Shared folder
- $0.10 / GB-month
- Traffic out
- $0.10 / GB
MACHINE · 24 GB
from$1.40/hr
NVIDIA L4
Image generation, light fine-tuning, most ComfyUI work
MACHINE · 48 GB
Popularfrom$2.64/hr
NVIDIA L40S
Video models, bigger LoRAs, 30B-class inference
MACHINE · 96 GB
from$4.74/hr
RTX PRO 6000 Blackwell
70B-class models, long video, what fits nowhere else
What a month costs
≈ $90/month
87 h of workspace $20.88 + 26 h on 48x32 $68.64. Disks and the shared folder are extra.
Credit is drawn at these rates, the same in the US and Europe.
Hinode vs RunPod
The same cards, with a computer around them
A 48 GB Hinode machine is the L40S RunPod rents, with your apps and your files already on it.
| Hinode | RunPod | Vast | Lambda | HF Pro | |
| GPU | L40S · 48 GB | L40S · 48 GB | RTX A6000 · 48 GB | A100 · 40 GB | ZeroGPU · shared |
| GPU access | Dedicated | Dedicated | Dedicated | Dedicated | Quota + queue |
| Interface | SSH, VS Code, AI assistants, apps in your browser | SSH / Jupyter | SSH / Jupyter | SSH / Jupyter | Spaces (web) |
| When you stop | Paused, files and apps kept | Volume kept | Tied to one host | Lost on terminate | Ephemeral |
| Devices | Mac, Windows, Linux and Android apps | Browser | Browser | Browser | Browser |
Swipe for Vast, Lambda and HF Pro →
FAQ
Frequently asked questions
A Linux computer with a graphics card, in the cloud. You keep a workspace with your files and apps, and add machines with a whole card when the work needs one.
The workspace is what you keep: a small slice of an NVIDIA L4, your home folder, the apps you install and the shared folder, at $0.24 an hour while it runs. A machine is a whole card, 24, 48 or 96 GB, that you attach beside it. It starts from your latest snapshot, so your apps are already there, and it bills by the minute until you pause or delete it.
Ubuntu with a working graphics card and a terminal. The rest is yours to pick: the app catalog installs ComfyUI, PyTorch, JupyterLab, Ollama, vLLM, VS Code and more in one click, and apt, pip and conda work as usual. Install on the workspace, save a snapshot, and every new machine starts with it.
By the minute, from prepaid credit added from $10. Eligible new accounts get $10 to try after a $1 card check and refund. The trial lasts up to 7 days; we stop before credit runs out and keep files for 24 hours after it ends. Paid credit never expires. The workspace is $0.24 an hour while it runs, machines $1.40 to $7.41. Disks are $0.15 per GB-month written, shared files $0.10 per GB-month, and traffic out $0.10 per GB.
Eligible new accounts get $10 after a $1 card check and refund, subject to the daily trial allowance. Try any GPU size on one machine for up to 7 days. About 4 hours 52 minutes on the smallest L4, before storage, traffic and the stop margin. Files are kept for 24 hours after the trial ends. Add prepaid credit from $10 to continue.
Only for storage. The workspace and every machine pause themselves after 15 minutes with no input, no GPU work, no traffic and none of your processes running, and their hourly charge stops. Disks keep everything and keep billing for what is on them. A job that is still running keeps its machine awake.
The Hinode app on macOS, Windows and Linux opens a machine's apps in your browser. SSH and VS Code work from any computer. The dashboard works in any browser and in the Android app: start a machine, check a job, read the bill.
Yes. ComfyUI runs on the cloud GPU, not on your Mac. The Hinode app for Mac opens it in your own browser. Your models and custom nodes are still there the next time you open it.
Yes. The workspace and every running machine answer over SSH at a name that does not change, so ssh, scp, rsync and VS Code work as with any server. The app sets up the key for you, and hinode run starts a machine and runs a command on it.
Yes. Claude lists Hinode in its connector directory, so adding it there takes one click. Anything else takes https://mcp.hinode.sh as a connector, and an assistant on your own computer runs hinode mcp. It can check prices, attach a machine, run a job and pause it, acting as you and refused whatever you would be refused.
Each workspace and machine has a home folder on its own disk, and the shared folder is one place every machine reads. Pausing keeps everything. Deleting a machine removes its own disk and leaves the shared folder alone; deleting the workspace removes everything in it. Disks are encrypted, and you can copy your data out at any time.
Set it up once. That is the last evening it costs you.
Create a workspace, install your apps, and start a machine when the work needs one.
Eligible new accounts get $10 to try after a $1 card check and refund.
