A private LLM inside your own cloud account.
The model runs in your AWS, Azure or Google Cloud account, in your region, behind your network controls and encrypted with your keys. No prompt or document goes to a model vendor.
What you get
Your account, your controls.
Deployed in your account
GPU capacity in your subscription or project, in the region you choose. You pay the cloud provider directly.
Private networking
Private endpoints only. No public API, and no outbound traffic by default.
Your keys
Storage and model weights encrypted with keys you hold in your KMS or HSM.
Logs where you want them
Requests, access and admin actions sent to your SIEM, under your retention rules.
Infrastructure as code
Terraform and Helm in your repository, so your team can review, rebuild and change it.
Runbook and recovery test
An operations runbook and a disaster-recovery test before go-live.
How it works
From landing zone to production.
01
Review
Your landing zone, network rules, identity and key management. We work to your cloud standards, not ours.
02
Size
Model, quantisation and GPU type chosen from measured throughput at your load. We check GPU quotas in your region early.
03
Deploy
Kubernetes with the GPU operator, the serving stack and private endpoints, all from code in your repository.
04
Test
Your evaluation suite, load tests and a recovery test, with the results written up.
05
Run
Your team runs it with our runbook, or we operate it with access granted per task.
Technical detail
What we install.
Platform
Managed Kubernetes with the NVIDIA GPU operator, private endpoints and your key management service.
EKS · AKS · GKE · PrivateLink · Private Service Connect
Serving
vLLM or SGLang for most workloads, TensorRT-LLM where peak throughput on NVIDIA hardware matters. A standard API in front, so your applications don’t change if the model does.
vLLM · SGLang · TensorRT-LLM
Sizing
A 70B-parameter model needs about 140 GB for weights at 16-bit, 70 GB at FP8 and 35–40 GB at 4-bit, plus memory for each conversation in flight. We size from measured throughput at your expected load, and re-run your tests after any quantisation.
FP8 · AWQ · GPTQ
Delivered with every deployment: Terraform / Helm · no outbound traffic by default · your keys (KMS / HSM) · logs to your SIEM · signed images and SBOM · runbook · disaster-recovery test
Who has access
No standing access for us.
Our engineers have no permanent access to your account. When we need to work on the system, access is granted just in time through your privileged-access tooling, and every session is logged by you.
Regulation
Cloud outsourcing rules, covered.
| Framework | What it asks | How we support it |
|---|---|---|
| MAS Outsourcing Guidelines (Banks), Notice 658 | Due diligence, contract terms and audit rights for outsourced services. | We accept audit and access rights for you and MAS, and support your materiality assessment. |
| MAS Technology Risk Management Guidelines | Controls for access, cryptography, resilience and third parties. | Our controls mapped to the relevant sections. |
| DORA Art. 28–30 | A register of ICT providers, exit strategies and mandatory contract terms. | Contract terms written for Art. 30, an exit plan and the data you need for your register. |
| EBA Outsourcing Guidelines | Governance of outsourced functions in EU banks. | Documentation for your outsourcing assessment. |
Questions
What cloud and security teams ask.
Does any data go to a model vendor?
No. The model is open-weight and runs on GPUs in your account. It makes no calls to an external model API.
Which clouds do you support?
AWS, Microsoft Azure and Google Cloud, in any region where you can get the GPU capacity you need.
Can we use GPU capacity we have already reserved?
Yes. We size the deployment to fit it and tell you if it won’t meet your load.
Who pays the cloud bill?
You do, directly to your provider. Our fees are separate, so you see exactly what the model costs to run.
What we don’t promise
Some claims we won’t make.
- GPU capacity in every region. Quotas and availability vary, so we check them before we commit to a date.
- That a cloud account is enough for every workload. If your data can’t go to a public cloud, choose your data centre or air-gapped.
Your cloud, your keys.
We’ll walk your cloud and security teams through the design.