GPU Cloud Documentation — API Reference & Guides

Introduction

Open Source AI Cloud is a UK-based GPU rental marketplace and AI infrastructure provider. We offer three core services:

GPU Instances — Rent NVIDIA GPU compute by the hour with full SSH access, starting from £0.30/hr.
GPU Hosting — List your idle NVIDIA GPU on our marketplace and earn 77% of every rental automatically.
Enterprise Clusters — Dedicated federated GPU clusters for startups, enterprises, and UK universities running LLM inference or training workloads.

All compute is UK-based with full GDPR compliance and UK data residency. We are 60% cheaper than equivalent AWS or Azure GPU instances.

Quick start — rent a GPU in 5 minutes

Create your accountGo to /dashboard and click Register. Enter your email and password.
Add a payment methodGo to /billing and add your card via Stripe. You only pay for what you use.
Launch an instanceGo to /instances and click Launch Instance. Choose your tier.
SSH into your containerCopy the SSH command from your instances page and connect from your terminal.
Start computingYour container has full GPU access. Install any framework and run your workloads.

terminal

$ ssh [email protected]

# Connected to NVIDIA Blackwell GPU · Instance #12

$ nvidia-smi

+---------------------------------------+

| NVIDIA Blackwell GPU 128GB VRAM CUDA 13.0 |

+---------------------------------------+

$ pip install torch && python3 -c "import torch; print(torch.cuda.is_available())"

True

Create an account

Go to /dashboard and click Register. You need a valid email address and password. No credit card is required to create an account.

You can explore the marketplace at /marketplace and try our free AI chat at /chat without an account.

Billing setup

Go to /billing to add a payment method. We use Stripe for secure card processing. Your card is charged per second while instances are running.

How billing works

Billing starts when your instance is launched and stops the moment you destroy it.
You are charged per second — so a 30 minute session at £0.30/hr costs £0.15.
Invoices are available in your dashboard.
Enterprise clusters are invoiced monthly.

Launch an instance

Go to /instances and click Launch Instance. Choose your instance type:

Type	Price	Best for
On-demand	£0.30/hr	Interactive work, experiments, short jobs
Interruptible	£0.21/hr	Batch jobs, training runs, long experiments
Inference API	£0.12/min	Model serving, API calls
Training API	£0.31/min	Fine-tuning, full training runs

Instances launch in under 60 seconds. Each instance is an isolated Docker container with full NVIDIA GPU access.

SSH access

Once your instance is running, connect via SSH using the command shown on your /instances page:

terminal

$ ssh gpu{INSTANCE_ID}@bastion.opensource-ai-cloud.uk

Replace {INSTANCE_ID} with your actual instance ID shown on the instances page.

We use Cloudflare Tunnel for SSH access. No VPN, no port forwarding, and no firewall configuration needed. Works from anywhere.

Using an SSH key

Your SSH public key is injected into the container automatically when you add it to your account settings. Alternatively, password authentication is available.

Instance users

Each instance gets a dedicated system user — gpu12 for instance #12, gpu13 for instance #13, and so on. You have full root access inside your container.

Docker containers

Every GPU instance runs as an isolated Docker container with full NVIDIA GPU passthrough via nvidia-docker.

What is pre-installed

CUDA 13.0 and cuDNN
Python 3.11 with pip
nvidia-smi and NVIDIA utilities
OpenSSH server
Ubuntu 24.04 base

Installing frameworks

install PyTorch

$ pip install torch torchvision torchaudio

install HuggingFace

$ pip install transformers accelerate datasets

install vLLM for inference

$ pip install vllm

Your container persists while the instance is running. Install packages once and they remain available for the session.

Destroy an instance

To stop billing, go to /instances and click Destroy. Billing stops immediately. The container and all data inside it are removed.

Save any work before destroying. Data inside the container is not persisted after destruction. Download results or push to Git before destroying.

Host a GPU — overview

If you have an idle NVIDIA GPU, you can list it on our marketplace and earn 77% of every rental automatically. Your GPU earns money while you are not using it, with payouts direct to your bank via Stripe.

77%

Your earnings

23%

Platform fee

Auto

Stripe payout

Host requirements

Requirement	Minimum	Notes
GPU VRAM	8GB	Any NVIDIA GPU with 8GB+ VRAM
Python	3.8+	Required to run the host agent
nvidia-smi	Any version	Must be installed and working
Docker	20.10+	With nvidia-docker runtime
Internet	10 Mbps upload	For SSH tunnel and heartbeat
OS	Ubuntu 20.04+	Other Linux distros may work

Accepted GPU models

Any NVIDIA GPU with 8GB+ VRAM is accepted, including:

RTX 3070, 3080, 3090, 3090 Ti
RTX 4070, 4080, 4090
A10, A100, H100, H800
NVIDIA Blackwell GPU, GB200
Tesla T4, V100

Host setup guide

Create an account and go to /hostRegister at /dashboard then go to /host.
Connect your Stripe accountClick Connect Stripe to link your bank account for automatic payouts.
Set your priceEnter your price per hour. We show you exactly what you earn (77%) and what the customer pays.
Download the host agentDownload host_agent.py from the host page or via curl.
Run the agentRun the agent on your GPU machine using your token.

Host agent

The host agent is a lightweight Python script that registers your GPU, sends heartbeats, and manages rental sessions.

download agent

$ curl -O https://opensource-ai-cloud.uk/host_agent.py

install dependency

$ pip install requests

run agent

$ python3 host_agent.py --token YOUR_TOKEN --price 1.50

run in background with screen

$ screen -S gpu-agent

$ python3 host_agent.py --token YOUR_TOKEN --price 1.50

# Press Ctrl+A then D to detach

Keep the agent running to stay listed on the marketplace. It sends a heartbeat every 60 seconds. If the agent stops, your GPU is automatically removed from the marketplace within 90 seconds.

Earnings and payouts

You earn 77% of every rental. Payouts go directly to your bank account via Stripe Connect.

Example rental	Amount
Customer pays (2 hours at £1.50/hr)	£3.00
Stripe processing fee	-£0.09
Platform commission (23%)	-£0.69
You receive (77%)	£2.22

First payout typically takes 2-7 days (standard Stripe schedule). Ongoing payouts are typically 2 days after each rental.

Enterprise GPU clusters — overview

We provide dedicated federated GPU clusters for businesses and institutions that need reliable, scalable AI compute. Our clusters pool multiple GPU hosts into a single high-availability resource with automatic failover, 99.9% SLA, and monthly invoicing.

We are significantly cheaper than AWS, Azure, or Google Cloud for equivalent GPU compute — typically 60% lower cost with better hardware availability.

Who is it for?

Customer type	Use case	Example
AI startups	LLM inference at scale, fine-tuning	Running inference for a B2B AI product
Scale-ups	Training proprietary models, batch inference	Weekly training runs for recommendation models
Enterprises	Private LLM deployment, RAG pipelines	Internal ChatGPT-style system on private data
UK universities	Research, student ML courses	Department-wide GPU access for AI degrees
Research institutions	Large model training, scientific compute	Climate modelling, protein folding, NLP research
Public sector	GDPR-compliant AI on UK data	NHS data analysis, government AI tools

Enterprise plans and pricing

Plan	Price	GPU nodes	VRAM	SLA	Support
Starter	£500/month	2 nodes	Up to 256GB	99.5%	Email, 24hr
University	£1,500/month	5 nodes	Up to 640GB	99.9%	Priority, 12hr
Research	Custom	10+ nodes	1TB+	99.99%	24/7 dedicated

All plans include UK data residency, automatic failover, monthly usage reports, and monthly invoicing. Submit an enquiry at /enterprise.

Automatic failover and SLA

Our federated cluster architecture ensures high availability through automatic failover:

Heartbeat monitoringEvery cluster node sends a heartbeat to our orchestrator every 30 seconds.
Failure detectionIf a node misses heartbeats for 90 seconds, the orchestrator marks it as offline.
Automatic failoverThe workload is migrated to a standby node automatically — no manual intervention needed.
Email notificationYour team receives an automatic email confirming the failover completed successfully.
SLA maintainedThe entire process completes in under 30 seconds, preserving your uptime SLA.

Build your own AI platform

We also help businesses build their own GPU cloud platform from scratch — a custom marketplace, billing system, SSH access, enterprise portal, and GPU orchestration layer built on the same open infrastructure powering Open Source AI Cloud.

Visit opensource-ai.co.uk to learn more about building your own platform. We handle everything from architecture to deployment.

What we can build for you

Custom GPU rental marketplace with your branding
Stripe Connect payment integration with custom commission splits
SSH access system via Cloudflare Tunnel
Host agent for GPU registration
Enterprise portal with SLA reporting
Federated cluster orchestration with automatic failover
Email notification system
API for programmatic access

API — authentication

The API base URL is https://api.opensource-ai-cloud.uk. All authenticated endpoints require a Bearer token in the Authorization header.

Login and get token

POST /auth/login

$ curl -X POST https://api.opensource-ai-cloud.uk/auth/login \

-H "Content-Type: application/json" \

-d '{"email":"[email protected]","password":"yourpassword"}'

response

{"access_token": "eyJ...", "token_type": "bearer"}

Register

POST /auth/register

$ curl -X POST https://api.opensource-ai-cloud.uk/auth/register \

-H "Content-Type: application/json" \

-d '{"email":"[email protected]","password":"yourpassword"}'

API — instances

Launch an instance

POST /instances/launch

$ curl -X POST https://api.opensource-ai-cloud.uk/instances/launch \

-H "Authorization: Bearer TOKEN" \

-H "Content-Type: application/json" \

-d '{"image_key":"cuda-12.3","tier":"ondemand"}'

List instances

GET /instances/list

$ curl https://api.opensource-ai-cloud.uk/instances/list \

-H "Authorization: Bearer TOKEN"

Destroy an instance

DELETE /instances/{id}/destroy

$ curl -X DELETE https://api.opensource-ai-cloud.uk/instances/12/destroy \

-H "Authorization: Bearer TOKEN"

Tier value	Price	Description
ondemand	£0.30/hr	On-demand instance
interruptible	£0.21/hr	Interruptible instance
inference	£0.12/min	Inference API
training	£0.31/min	Training API

API — marketplace

List available GPUs

GET /marketplace/gpus

$ curl https://api.opensource-ai-cloud.uk/marketplace/gpus

response

[{"id":1,"gpu_name":"NVIDIA Blackwell GPU","gpu_vram_gb":128,"price_per_hour":1.50,"available":true}]

API — host

Register as a host

POST /host/register

$ curl -X POST https://api.opensource-ai-cloud.uk/host/register \

-H "Authorization: Bearer TOKEN"

Set price

POST /host/set-price

$ curl -X POST https://api.opensource-ai-cloud.uk/host/set-price \

-H "Authorization: Bearer TOKEN" \

-H "Content-Type: application/json" \

-d '{"price_per_hour": 1.50}'

Host agent registration

POST /agent/register

# Called automatically by host_agent.py every 60 seconds

$ curl -X POST https://api.opensource-ai-cloud.uk/agent/register \

-H "Content-Type: application/json" \

-d '{"token":"HOST_TOKEN","gpu_name":"NVIDIA RTX 4090","gpu_vram_gb":24,"price_per_hour":0.80}'

API — enterprise clusters

List clusters

GET /clusters/list

$ curl https://api.opensource-ai-cloud.uk/clusters/list \

-H "Authorization: Bearer TOKEN"

Get cluster status

GET /clusters/{id}/status

$ curl https://api.opensource-ai-cloud.uk/clusters/1/status

Enterprise enquiry

POST /enterprise/enquiry

$ curl -X POST https://api.opensource-ai-cloud.uk/enterprise/enquiry \

-H "Content-Type: application/json" \

-d '{"name":"Jane Smith","institution":"Acme AI","email":"[email protected]","gpu_count":5,"requirements":"LLM inference cluster"}'

NVIDIA Blackwell GPU

Specification	Value
GPU model	NVIDIA Blackwell GPU
Platform	NVIDIA Blackwell GPU (DGX Spark)
Architecture	Grace-Blackwell SoC
VRAM	128GB unified memory
CUDA version	13.0
Driver version	580.142
CPU	ARM64 Grace
OS	Ubuntu 24.04
Location	United Kingdom

Supported host GPUs

Any NVIDIA GPU with 8GB+ VRAM can be listed on our marketplace. Here are commonly listed models:

GPU model	VRAM	Typical price	Best for
NVIDIA Blackwell GPU	128GB	£1.50/hr	Large model training, 70B+ inference
NVIDIA H100	80GB	£2.00/hr	Fast training, transformer workloads
NVIDIA A100	80GB	£1.20/hr	Training, large batch inference
NVIDIA RTX 4090	24GB	£0.60/hr	Fine-tuning, smaller models
NVIDIA RTX 3090	24GB	£0.40/hr	Experiments, development
NVIDIA RTX 3080	10GB	£0.25/hr	Small models, inference

View all currently available GPUs at /marketplace.