Approved cloud
Requests cross to allowlisted external models, after the boundary has transformed them.
Designed for teams already using frontier models under contract.
Deployment
Where controlled execution runs
Controlled execution is designed to span approved cloud, private cloud/VPC, on-premises, edge, and isolated environments, chosen per workflow, per environment, per pilot. Not every product supports every environment today; the matrix below says which.
Built for where you operate
Data residency by design: the Gulf, the EU, and any estate where information may not leave.
- Gulf data residency
- EU data residency
- Fully isolated estates
Deployment patterns
Runs where your constraints live.
The same boundary architecture is designed to deploy from approved cloud through fully air-gapped environments. Pilot deployments are scoped to your perimeter.
Requests cross to allowlisted external models, after the boundary has transformed them.
Designed for teams already using frontier models under contract.
Portfolio matrix
Where each product is intended to run.
Per-cell status uses the same vocabulary as the product pages: "Available in pilots" cells are configured for your environment during the pilot; "Designed" marks an intended pattern, not a package. Boundary governs deployment constraints across every row.
| Product | Approved cloud | Private cloud / VPC | On-prem / edge | Isolated / air-gapped |
|---|---|---|---|---|
| Boundary | Designed | Designed | Designed | Demonstrated (synthetic Local-Only policy model) |
| Cortex Assist | Designed | Available in pilots | Available in pilots | Designed |
| Cortex Extract | Designed | Available in pilots | Available in pilots | Designed |
| Cortex Voice | Not offered | Designed | Available in pilots (local processing) | Designed for isolation |
| Forge | Designed | Designed | Available in pilots (training) | Designed |
| AxonCX | Available in pilots | Designed | Designed | Not suited (external channel connectors) |
| Axon Live Agent | In development | In development | Designed | Not planned |
Forge is especially relevant to local, on-premises and edge execution; Cortex Voice is designed around local processing; Axon's external-channel connectors are not suited to true air-gap operation.
In practice
Notes per pattern.
Private cloud / VPC
The Boundary and approved models can run inside your own cloud tenancy, with data paths, dependencies, and operational controls defined during deployment design.
Scoped for regulated enterprises during pilot deployment.
On-premises / edge
Designed for deployment on customer-controlled infrastructure, next to the systems it protects, scoped and integrated during the pilot.
Intended for restricted facilities and field conditions.
Air-gapped
Designed for environments where no external route exists: local models only, with the audit trail kept inside the perimeter.
The Local-Only demonstration illustrates the intended contained-routing behaviour using synthetic data.
Core capabilities already exist. Deployment packaging, infrastructure integration, model selection, and operational controls are configured for each pilot environment; none of these patterns is an off-the-shelf package today.
Regional constraints
Controlled execution in Gulf environments
Kordane is developed from Dubai and designed to support customer-defined residency, policy and audit requirements. It is configured around your legal, security and residency constraints; it does not make compliance claims on your behalf.
- May this information leave the UAE?
- Environment policy decides per information class. Residency rules are explicit, not implied.
- Which model endpoints are permitted?
- Only allowlisted routes are reachable; anything else is refused and recorded.
- Must tokenisation happen locally?
- Transformation runs inside your perimeter before any crossing.
- Where does decision evidence live?
- Retention location and rules are customer-defined during pilot scoping.
- Can frontier models be used at all?
- Through permitted routes, after transformation, and only when the task requires them.
- How are Arabic tasks evaluated?
- Evaluation sets are task- and language-specific, agreed during the pilot.