Deployment modes compared
RSEEN is built so model traffic can be routed through different deployment postures without changing the document pipeline. The modes below are intentionally described by control boundary rather than by private provider names.
Modes at a glance
| Mode | Status | Where model inference runs | Cross-border posture | Who operates it |
|---|---|---|---|---|
| Sovereign hosted gateway | Supported | Saudi-hosted contracted infrastructure | Intended to stay in Kingdom; still an external dependency | Gateway provider, governed by customer contract |
| Operator-controlled endpoint | Supported | Customer-controlled endpoint or approved infrastructure | Depends on endpoint location; can be fully on-prem | Customer or approved infrastructure partner |
| Saudi cloud-GPU deployment | Planned | Saudi-region GPU infrastructure | Intended to stay in Kingdom | GulfBoost-managed |
| On-prem hardware | Planned | Customer data centre | No external inference path when configured that way | GulfBoost-managed on customer premises |
Sovereign Hosted Gateway
This posture uses a Saudi-hosted model gateway selected through the admin configuration surface. It can be the right fit when an institution accepts a contracted external provider as long as inference remains inside the Kingdom and the provider relationship is governed appropriately.
The important caveat: a sovereign hosted gateway is still an external dependency. It has its own operator, contract, availability profile, and due-diligence burden.
Operator-Controlled Endpoint
This posture sends model calls to an OpenAI-compatible endpoint controlled by the operator or a specifically approved infrastructure partner. If that endpoint runs inside the customer’s own network perimeter, inference traffic does not need to leave that perimeter.
This is the stricter control posture for institutions that want no external model dependency at inference time. It shifts responsibility to the operator: hardware sizing, model selection, endpoint availability, and operational monitoring.
Planned Modes
Saudi cloud-GPU and on-prem hardware deployments are product directions for customers that want GulfBoost to operate more of the stack. Until those paths are fully wired, treat them as deployment positioning rather than a documented install procedure.
How to Choose
Start with policy, not convenience. If the institution requires no external model service, use an operator-controlled endpoint inside the perimeter. If the institution permits a Saudi-hosted external provider under contract, a sovereign hosted gateway may be operationally simpler. If neither fits, the planned managed options are the right commercial discussion.