Documentation

How to rent an instance, host a model, and talk to the inference API. If something here is wrong, tell us; the docs are maintained by the same four people who run the hardware.

What is GPUs4All

Two products on the same account. GPU instances are raw machines you control over SSH. Model hosting takes a checkpoint and gives you an OpenAI-compatible endpoint, which is most of what people want the GPU for in the first place.

Choosing the right product

	GPU instances	Model hosting
You get	a machine and an IP	a URL and a key
You control	everything	model and scaling settings
Billing	per second while running	monthly base + per token
Good for	training, custom stacks, anything unusual	serving, evaluation, product integrations

Account basics

• New accounts start with $10 of credit, no card required.
• One key per account for the inference API. Rotate it in Settings.
• Every deployment and instance has a spending cap. We stop work at the cap instead of billing past it.
• You can leave any time. Data is yours, exportable, and deleted when you ask.

Next steps

Start with the quickstart for a working first request, then read the API reference when you need the exact fields.