Alongside the three pillars
Artificial Intelligence
AI on your own premises: on your platform, behind your access rules, every request logged audit-proof.
Back to the overall concept
Your business units already use AI, mostly through external providers, and no one in the company can say which data went where. For a regulated company that cuts twice: data protection can hardly approve transfers to third-party services, and the EU AI Act classifies risk assessment and pricing in life and health insurance as high-risk. Only the date moved, to December 2027, not the requirements. The transparency duties apply from August 2026.
The way out is AI in-house, and that is where it gets stuck: experience with which use cases carry their weight is missing, and purchased compute sits idle because it is dedicated to a single team.
We make in-house AI work: from the first use case through model operation to logging that satisfies your evidence duties. Dedicated GPUs become a shared pool with guaranteed shares.
What changes for you
Your teams use AI through an approved in-house path, the drift to external services ends
Outbound connections to external model providers are blocked at the network, your data does not leave
Every request carries user, model and version, the documentation duty fulfills itself in operation
Several teams share the same GPUs with guaranteed shares, expensive reservations stop sitting idle
M1: GPUs and allocation
Base for M2 and M3
Driver and device management through the node lifecycle, no manual fixing after updates
Splitting one GPU across several workloads, hard-partitioned or time-shared, depending on hardware
Fixed shares per tenant instead of a race for free GPUs
Scheduling separates training from inference, response times stay stable
Separated from other workloads, expensive GPUs stay reserved for what they were bought for
Utilization visible per GPU, idle reservations get flagged
M2: Inference and models
Module
Endpoints scale with load and shut down when idle
A model registry with version, provenance and approval status
Weights in your own storage, checksum on load
An interface compatible with the market standard, moving off an external provider requires no changes to your applications
Models ship through the same delivery path as the rest of the platform
M3: Access and logging
Module
Access through the same identity integration as the rest of the platform
Every request stored tamper-proof with user, model, version and timestamp
Outbound connections to external model providers blocked at the network level
Filters for inputs and outputs where the business defines them
Costs per GPU, tenant and endpoint
Documentation toward Articles 12 and 13 of the EU AI Act
Builds on the platform pillar, and also runs on your existing platform once it has passed our audit. You provide: GPUs and a contact in data protection. Use cases, procurement and model selection take shape in the workshop.
Run AI accountably
Transparency duties of the AI Act under Article 50, in force since August 2026.
Every request recorded with user, model and version, documentation for Articles 12 and 13.
The full comparison, all seven duties