Run your sensitive ML workloads on EU soil. Zero DevOps.
GhostNexus is a Python GPU execution API hosted in Frankfurt, built for healthtech and regulated teams that need to process ML workloads inside the European Union — with a GDPR Art.28 DPA available within 24 h.
For: CTOs · DPOs · Technical founders · Heads of Data / AI
Your ML pipeline may have a GDPR blind spot.
You use an international GPU platform to run your models. It works. The problem: international GPU platforms can involve region, sub-processing or transfer choices that are hard to document for a regulated team.
Data processing agreement
The GPU platform does not provide a GDPR Art.28 DPA — your DPO blocks the deployment.
Data location
Impossible to confirm in writing that data never left the EU during execution.
In-house GPU infra
Building compliant GPU infrastructure internally: 3 to 6 months. Meanwhile, the model waits.
GhostNexus reduces this to 24 hours. You submit your use case, you receive the DPA and pilot access. Your workloads run in the EU without you managing any infrastructure.
Managed GPU execution. Data in Europe. GDPR documentation included.
We host the GPUs. We manage the infrastructure. We provide the Art.28 DPA.
import requests
response = requests.post(
"https://api.ghostnexus.net/api/jobs",
headers={"x-api-key": "YOUR_KEY"},
json={
"task_name": "inference-pipeline",
"script_content": open("my_model.py").read(),
}
)
# → Runs on a Frankfurt GPU · Data deleted after the job- 1Account + API key
- 2Submit script + data
- 3GPU execution in Frankfurt
- 4Retrieve results
- 5Data deleted
What you don't have to manage
- No server to provision
- No CUDA, Docker or Kubernetes setup
- No data processing agreement to draft from scratch
- No hosting audit to initiate
What you get immediately
- GDPR Art.28 DPA signed by GhostNexus
- Written confirmation of processing location (Frankfurt, DE)
- Documented REST API
- Concierge onboarding for early customers
- Direct support — response within 24 business hours
- DPA and pilot access within 24 h after use-case validation
The hardware, precisely.
You can't evaluate a GPU offer without the specs. Here they are.
- · 20 GB GDDR6 ECC VRAM
- · Dedicated server — never shared between clients
- · Hetzner datacenter, Frankfurt, Germany
- · CUDA 12, PyTorch / HuggingFace preinstalled
- · LoRA / QLoRA fine-tuning of 7-8B models (LLaMA 3, Mistral)
- · Inference of models up to ~13B quantized
- · Embeddings, classification, medical vision (CNN, ViT)
- · Batch pipelines: scoring, clinical NLP, image processing
Need more (70B, multi-GPU)? Extended capacity on quote — same Frankfurt location, same DPA.
We don't publish theoretical benchmarks copied from vendor datasheets. Measured execution times on your real use case are provided during onboarding, before any commitment beyond the first month.
What GhostNexus is not.
Clarity avoids bad surprises — especially for your DPO.
Note: if your use case involves long-term storage of personal health data or strictly requires HDS certification, we will tell you before starting. Some cases do not fit our offer.
Paid pilot. Five seats. No commitment.
No mandatory sales call. DPA and access provided within 24 h after use-case validation.
No commitment · Cancel at day 30
- Full API access
- Hetzner Frankfurt GPU
- Signed GDPR Art.28 DPA
- Concierge onboarding
- Direct support — response within 24 h
- Usage volume defined at onboarding
From the 6th customer
- Everything in the pilot
- Extended volume
- Negotiable SLA
- Monthly invoicing
What we state — and what we don't promise.
We are direct about this because your DPO will be too.
Processing location
Workloads submitted via the API run exclusively on Hetzner servers, Frankfurt datacenter, Germany — European Union.
GDPR Art.28 DPA
Provided within 24 h, signed by GhostNexus, based on GDPR requirements and informed by CNIL-recommended clauses.
Data retention
Data submitted for execution is deleted after each job. Logs are kept 30 days and deletable on written request.
Sub-processors in the execution chain
No non-EU sub-processor is involved in executing your ML workloads. Billing, email and support services are documented separately.
Data access
No human access to workload data by default, unless explicitly requested by the client for support or debugging.
What we do not claim
- We are not HDS certified. If your case strictly requires it, we will tell you before any commitment.
- We do not replace your DPIA. We can provide the technical elements you need to complete it.
- We do not guarantee absolute compliance of your processing — that depends on your overall architecture.
GhostNexus is designed to help regulated teams reduce exposure to risks linked to processing sensitive data outside the European Union. This text does not constitute legal advice.
“What if you disappear?” — fair question.
GhostNexus is an independent French business. You have every right to ask what happens if it stops. Here are the commitments, in writing.
Zero technical lock-in
Your scripts are standard Python (PyTorch, HuggingFace). No mandatory proprietary SDK. Migrating to OVH or in-house infra = changing one API URL.
Guaranteed reversibility
Export your data and results at any time, free of charge. If the service ever shuts down: 60-day written notice and migration assistance, in the contract.
Nothing to recover
Workload data is deleted after each job. In practice there is no data stockpile at GhostNexus you could be held hostage over.
Live service status: ghostnexus.net/status — monitored automatically every 30 minutes.
Frequently asked questions
Do workloads stay in Europe?
Are you HDS certified?
How does the DPA work?
What happens after the pilot?
Can we use our own models?
Why a flat monthly rate instead of usage-based billing?
Request a pilot — €299/month
5 seats at the founding rate. DPA and access provided within 24 h after use-case validation. No commitment.