Your data never leaves your infrastructure
MLVisor runs inside your environment. There is no external LLM, model or inference API in the loop. Below is exactly where the boundary is drawn.
No external model in the loop
VisorEngine is our own probabilistic inference engine. It does not wrap someone else's LLM and it makes no calls to a third-party inference API. Your data is never sent to an outside provider to be analysed — because there is no provider to send it to.
Runs where your data lives
MLVisor is deployed into your environment: your own hardware, your private cloud, or an isolated network. Your datasets stay in your storage and processing happens on your compute.
You own the model weights
Trained models and their weights are produced inside your deployment and stay there. They do not leave your cluster without your explicit consent, and MLVisor neither copies nor uses them for its own purposes.
What stays, what leaves
Two explicit lists instead of a general assurance — something your security team can check line by line.
Stays inside your cluster
- Your raw datasets and everything derived from them
- Feature-engineering outputs and intermediate tables
- Trained model weights and checkpoints
- Your queries, prompts and job definitions
- Generated code, interfaces and deployment reports
Leaves only if you allow it
- Licence validation — deployment identifier and seat count
- Product telemetry and error reports, when you explicitly enable them
- Diagnostic bundles you choose to share when you open a support request
- Datasets or models you choose to publish to VisorHub
None of these carry the contents of your datasets. All of them can be switched off, and in an air-gapped deployment they are off by default.
Platform security controls
The controls that sit on top of the architecture, for day-to-day operation.
Encryption
AES-256 at rest, TLS in transit. Encryption keys stay inside your environment.
Access control
Role-based permissions with granular authorisation at the project, dataset and model level.
Network isolation
Can operate with no outbound internet access. Air-gapped deployment is supported.
Audit trail
Who accessed which dataset, which model was trained, what was deployed — all recorded.
Reproducibility
Every run is retained with the code stack and metrics it produced, so a result can be verified later.
Supply chain
The inference layer is our own code — no external model vendor, and therefore no supply-chain exposure on that surface.
Where you can run it
On your own hardware
Installed on your bare-metal servers. Nothing travels outside your data centre.
In your private cloud
Runs in your own cloud account, inside your own network and identity provider.
Air-gapped
Offline installation and licensing for isolated environments with no internet connection.
Exactly where we stand today
MLVisor is a pre-launch product. Rather than imply a certification we do not hold, here is the position as it actually is.
Data residency
In placeTrue today — data does not leave your infrastructure.
KVKK disclosure obligations
In placeTrue today — met by our privacy and cookie policies.
GDPR alignment principles
In placeTrue today — data minimisation and retention limits are applied.
Independent penetration test
PlannedPlanned before commercial launch (Q1 2027).
SOC 2 Type II
PlannedTargeted after commercial launch. We do not hold this certification today.
ISO/IEC 27001
PlannedTargeted during the scaling phase. We do not hold this certification today.
Found a vulnerability?
We take responsible disclosure seriously. Write to us with “Security disclosure” in the subject line and we will respond within 72 hours. Please hold public disclosure until a fix has shipped.
[email protected]Running a security review?
For architecture documentation, deployment topology and vendor-assessment questionnaires, write to the team directly. We answer enterprise reviews in detail.