Apps on rented GPUs
One-click apps on a cloud GPU
Rent a GPU by the minute and it starts with the app installed, running and behind its own private link: ComfyUI for images and video, Open WebUI with Ollama for private chat, Unsloth Studio for fine-tuning. Same price as the clean machine.
ComfyUI
Image and video generation
- Node workflows for Flux, SDXL, Wan and Z-Image
- Download to this GPU: a workflow's models straight onto the machine
- ComfyUI-Manager on
Open WebUI
Private chat with open models
- Ollama on the GPU: pull Llama, Qwen, Gemma, DeepSeek, Mistral
- Ollama's API on localhost:11434, over SSH
- Chats stay on the machine
Unsloth Studio
Fine-tune open models
- Train Llama, Qwen, Gemma on your own data
- Chat with your model, export to GGUF
- Always Unsloth's latest release, with its own sign-in
Or the clean machine: Python, PyTorch and CUDA ready, install anything with pip, conda or apt. Browse GPUs
How it works
From Rent to an open app
- 01
Pick a GPU and an app
Filter the list by the app you want: only the machines with enough VRAM and disk for it are shown. Its card says what it needs.
- 02
It installs on its own
The machine answers first (JupyterLab or SSH, billing starts there); the app sets up on top, at a version we tested, and the start screen shows each step.
- 03
Open it from its own link
Every app gets a private address with a token; nobody else can open it and no port is open on the provider's network. Apps with a login get a password for that rental.
FAQ
Questions
What is a GPU app (or template)?
A rental that starts with an application already installed and running, instead of a bare machine: you choose ComfyUI, Open WebUI with Ollama or Unsloth Studio at Rent, and open it from its own link a few minutes later. Other clouds call these templates.
Does an app cost extra?
No. A rental with an app costs the same as the clean machine, billed per minute from the moment the machine answers. The install runs on that time, usually a minute or a few, at the machine's network speed.
Which GPU do I need?
Each app has a minimum per GPU: 16 GB of VRAM for ComfyUI, 8 GB for Open WebUI with Ollama, 12 GB for Unsloth Studio. The list filtered by the app only shows machines that meet it; the guides say which model fits which card.
Are my models and files kept after the rental?
No. Everything on the machine is deleted when the rental ends, models included. Download what you want to keep, or push it to Hugging Face, before you stop.
Can I still use the terminal?
Yes. It is a normal rental with the app on top: JupyterLab or SSH, root, pip and nvidia-smi as on the clean machine.
Can I start an app from code or an AI agent?
Yes. Add "template": "comfyui", "openwebui" or "unsloth" to POST /api/v1/rentals, or ask an agent with the ParalonCloud MCP server: the rental reports the install and gives the app's link, and its sign-in when it has one.