AI on your infrastructure
Three deployment modes, anonymised personal data, DIFC DPL and GDPR: what you bring to your security team before the project starts.
Three deployment modes
On your servers, in our cloud in your region, or a fully air-gapped setup on open-weight models on request.
Personal data
Anonymised before it reaches the AI, never used for training, every model call logged.
Legal: DIFC DPL and GDPR
DIFC Data Protection Law, GDPR where applicable, Numisnipe Tech Ltd. Security documentation before the project starts.
Where the AI runs: three deployment modes
| Your servers | Our cloud in your region | Air-gapped setup | |
|---|---|---|---|
| Where the model runs | Inside your perimeter, deployed by our team | Cloud region of the client’s choice, a dedicated environment | Fully isolated installation on request |
| Which models | Open-weight models or your licences | Open-weight and commercial models by agreement | Open-weight models only, in your environment |
| What happens to data | Never leave your perimeter | Personal data is anonymised before it reaches the model | Not a single outbound call |
What happens to personal data
- Anonymised before it reaches the model.
- Never used to train models.
- Every model call is logged.
- You decide retention periods and log access.
Legal
DIFC Data Protection Law, GDPR where applicable, Numisnipe Tech Ltd (licence CL6629). Data residency of the client’s choice.
Pack for your security team
We send it before the project starts. Inside:
- Architecture and data flows.
- Threat model and safeguards.
- Access policy and logging.
- Legal documents and company details.
Questions security teams ask
Which models do you use?
Open-weight models in your environment or in our cloud; commercial models only by agreement and without personal data.
Where do prompts and responses go?
Only to the model in the chosen deployment mode. Personal data is anonymised before it is sent, every call is logged, and in the air-gapped setup nothing leaves your environment.
Who can access the log, and how do you revoke an agent’s access?
The log is available to your executive and your security team. Any agent can be stopped in one click, and access is revoked under your own procedure.
Agents under control.
Control is the product. Everything the agents do is visible, revocable and measurable.
The executive console shows every agent action, its permissions and its outcome. Disputed cases go to a human decision queue. Any agent can be switched off in one click.
Audit trail
Every agent action in a single log: who did what and on what grounds.
Permissions and rights
Each role's area of responsibility is set explicitly. An agent never exceeds its permissions.
Human review before action
Disputed and risky cases go to a human for a decision, before the action, not after.
Stop any agent in one click
Any agent can be paused or switched off in one click.
Before and after
We measure before and show after. The result is visible in numbers, not in impressions.
Start with a free 15–30 minute audit
We will identify the priority business process, show where an agent delivers a measurable effect and name the launch timeline for the first agent.
Pick a time in the calendaror leave a request