Private AI Pilot
Prove it on your workload before you commit.
Validate your workload on dedicated infrastructure before committing to a larger deployment. Deploy your model on dedicated capacity, run your real traffic against it, and decide with measured results instead of assumptions.
Pilot includes
- Model deployment on dedicated capacity
- Private OpenAI-compatible endpoint
- Benchmarking on your real workload
- Architecture recommendation
- Migration plan
Pricing
Contact sales
Scoped to your workload and quoted in writing up front.
Why a pilot
The cheapest infrastructure decision is the one you validate first.
Model choice, quantization and context length decide whether a deployment works — and they interact in ways a spec sheet can't predict. A pilot answers those questions with your prompts, your documents and your traffic pattern.
For teams moving off a public API
Check that an open-weight model meets your quality bar before you rewrite prompts for production.
For teams with sensitive data
Validate a private, single-tenant deployment with configurable logging before involving real data.
For larger rollouts
Size a multi-node or high-availability architecture from measurements, not vendor estimates.
For agencies
Prove a client use case on dedicated capacity before you quote the project.
What's included
Five deliverables, all of them yours to keep.
Model deployment on dedicated capacity
Private OpenAI-compatible endpoint
Benchmarking on your real workload
Architecture recommendation
Migration plan
How a pilot runs
Six steps from first call to a decision.
Duration depends on your workload and how quickly your team can send test traffic. We agree the timeline in the scoping call.
- 1
Scoping call
A technical call with an engineer: your use case, the model you want to test, data classification, connectivity and what the pilot must prove.
Output: Written pilot scope, success criteria and quote
- 2
Deployment
We deploy your model on dedicated 96 GB capacity, configure quantization and context length, and issue keys for a private endpoint.
Output: Private OpenAI-compatible endpoint
- 3
Benchmark on your workload
Your team sends representative traffic. We measure latency, throughput, memory headroom and output quality against the criteria you defined.
Output: Benchmark report with methodology
- 4
Architecture recommendation
Shared API, a single managed node, multiple nodes or high availability — sized from measured numbers, not guesses.
Output: Recommended architecture and monthly price
- 5
Migration plan
How to move from your current provider or prototype: endpoint changes, prompt adjustments, rollout steps and fallbacks.
Output: Step-by-step migration plan
- 6
Decision
You decide whether to proceed, with the evidence in hand. If the workload is the wrong fit for our infrastructure, we will tell you plainly.
Output: Go / no-go, on your terms
Success criteria you define
Before deployment, we agree in writing what “success” means for your workload. Typical criteria:
- Answer quality on your own evaluation set
- Latency at your real prompt and response lengths
- Throughput at your expected concurrency
- Context length your documents actually need
- Structured output or tool-calling reliability
- Effort to integrate with your existing OpenAI-based code
What we need from you
Pilots move as fast as test traffic arrives. The more representative your inputs, the more useful the results.
- A technical contact who can run test traffic
- Representative prompts or documents — synthetic or anonymised where possible
- The model you want to test, or the task if you're unsure
- Success criteria you are willing to put in writing
- Expected volume and concurrency for production
- Internal sign-off if the pilot involves personal data
Apply
Tell us what the pilot should prove.
An engineer reviews every application and replies by email to schedule a scoping call. Pilot capacity is dedicated, so we plan start dates with you.
Already sure about dedicated capacity?
Skip the pilot and deploy a Managed AI Node directly — from €699 per month.
Deploy Managed AIFAQ
Pilot questions.
Is the pilot free?
Pilots are scoped and priced individually, because the effort depends on the model, the workload and the connectivity you need. You receive a written quote with the scope before anything is deployed.
What happens after the pilot?
You get a benchmark report, an architecture recommendation and a migration plan. If you go ahead, we move you onto the recommended plan. If you don't, you keep the report.
Can the pilot use personal data?
Where possible, run the pilot on synthetic or anonymised data. If personal data is required, we put a Data Processing Agreement in place first, and you should review international-transfer requirements internally, since compute currently runs outside the EEA.
How fast can deployment happen?
Managed AI API access can be issued shortly after sign-up. Dedicated nodes are deployed after a short technical call to confirm model, context length and connectivity; timing depends on current capacity and your requirements, and we confirm it before you commit.
Can I bring or run my own model?
On Private AI Nodes, yes — including fine-tuned variants of supported architectures — provided the model is technically compatible with Apple Silicon inference runtimes, fits in memory with your required context length, and you have the rights to deploy it. We check compatibility before deployment.
Can you deploy 70B models?
Some 70B-class models can run in quantized form (typically 4-bit) on a single 96 GB node. Usable context length is constrained by the memory left for the KV cache. We validate quality and performance on your workload before committing.
Do you support CUDA workloads?
No. Our current infrastructure is Apple Silicon, which does not run NVIDIA CUDA. Workloads that depend on CUDA kernels, TensorRT or CUDA-only libraries — such as many training pipelines — need a different architecture, and we will tell you honestly if that applies to you.
Can European companies use the service?
Yes. European businesses may use infrastructure outside the EEA. Workloads involving personal data may require appropriate international-transfer safeguards and internal legal review. We support contractual and technical measures such as a Data Processing Agreement and Standard Contractual Clauses where applicable. Speak with us about your data classification before deployment.
Decide with evidence, not with a slide deck.
Bring the model and the workload. Leave with measured results, an architecture recommendation and a migration plan.