Skip to main content
No servers · No Docker · No idle billing

Serverless GPU API
for Python developers

Submit a .py script. We run it on a GPU. You pay per second of actual compute — nothing else. No instance to spin up, no Docker image to build, no idle time charged.

How it works

01

Write your Python script

Any standard Python script. Install packages via pip in the script itself. No Docker, no container, no environment setup.

# train.py import subprocess, sys subprocess.check_call([sys.executable, "-m", "pip", "install", "-q", "torch"]) import torch device = torch.device("cuda") # ... your training code
02

Submit via API or SDK

POST your script to our REST API or use the Python SDK. The job queues instantly and starts on the next available GPU.

from ghostnexus import Client client = Client() # uses GHOSTNEXUS_API_KEY env var job = client.run(open("train.py").read()) print(job.id) # job_abc123
03

Stream logs in real time

WebSocket log streaming lets you watch stdout/stderr as the job runs. Or poll the status endpoint.

# Stream live output job.stream_logs() # Or check status status = client.status(job.id) print(status.state) # running | success | failed

vs Traditional GPU Cloud

FeatureGhostNexus (serverless)Traditional GPU cloud
Instance managementNone — fully managedYou manage pods/instances
BillingPer second — zero idle costPer hour (billed even idle)
Setup time< 1 minute5–20 min (Docker, config)
Python onlyYes — submit a .py fileRequires container image
Log streamingReal-time WebSocketVaries by provider
CI/CD readyGitHub Action built-inManual scripting

Use cases

🧠

Model fine-tuning

Fine-tune LLaMA, Mistral, or any HuggingFace model. Submit the training script, get the saved checkpoint back.

🖼️

Stable Diffusion batch

Generate hundreds of images in a single batch job. No per-image API cost — pay only for GPU time.

📊

Batch inference

Run predictions on a dataset. Submit once, get all results. Ideal for nightly pipelines or data enrichment.

🔬

Research experiments

Run ablation studies or hyperparameter sweeps from CI/CD. Each experiment is a separate GPU job.

🏥

GDPR-sensitive workloads

EU-hosted, Docker-isolated per job. Signed DPA available. Safe for medical, financial, or biometric data.

Real-time inference API

Trigger GPU jobs from your backend via REST. Results available via webhook or polling within seconds of completion.

Built for CI/CD pipelines

The official GitHub Action lets you trigger GPU jobs directly from your workflow. Run a training job on every push to main, or trigger batch inference on a schedule.

- name: Run GPU job uses: Milaskinger/ghostnexus-run@v1 with: api_key: ${{ secrets.GHOSTNEXUS_API_KEY }} script: train.py stream_logs: "true"
See all integrations

Simple per-second pricing

No subscription. No minimum. No idle billing.

RTX 4090
$0.50/hr
Consumer powerhouse
A100 80GB
$2.20/hr
Data center grade
H100 80GB
$3.50/hr
Fastest available
Billed per secondNo idle time charged$5 free credits on signupNo credit card for trial

Run your first GPU job in 2 minutes

Sign up, get $5 free credits, and submit your Python script. No Docker, no infra setup.