Models
Selected after testing
Sovereign on-premise AI · Muscat, Oman
On-premise AI for institutions in Oman, scoped to the workload and deployed at the customer's site.
Selected open models can be evaluated, an air-gapped option is available, and Arabic and English requirements are documented before implementation.
Built for institutions that cannot use shared AI infrastructure
Public institutions
Regulated enterprises
Healthcare
Research
Professional services
Archives and knowledge teams
These are workload categories, not customer references. No customer identities are published.
01 · The product
Three boundaries documented before implementation.
I.
Customer-site deployment.
The location, network routes, storage, identity and administrative boundary are reviewed with the institution and recorded in the delivery scope.
Customer-site option
Air-gap option
Documented network boundary
II.
Arabic and English evaluation.
Language behaviour is tested on representative, approved material before the model and acceptance criteria are confirmed.
Representative Arabic samples
English workflow samples
Written acceptance criteria
III.
Written delivery terms.
Commercial terms are written into the quotation and deployment contract. Optional maintenance and update support are scoped separately.
Terms documented before purchase
Documented update path
Optional maintenance scope
A note to our readers
حدودُ البيانات الواضحة تبدأُ من نطاقٍ مكتوب.
A clear data boundary begins with a written scope.
Arabic and English requirements are reviewed as part of discovery.
02 · How it works
Five planning layers. One written scope.
01 · Kernel
Selected model family
Workload validation
Runtime selected after testing
Context sized by model
02 · Retrieval
Approved corpus
Language requirements
Ingestion scope
Retrieval evaluation
03 · Admin
Identity requirements
Local roles
Policy mapping
Data classification
04 · Audit
Event requirements
Retention boundary
Export destination
Acceptance evidence
05 · Air-gap
Air-gap option
Required routes mapped
Update procedure
Security review
03 · Configuration planning
Three starting points. Every scope confirmed by quotation.
Kernel
النواة
A single-appliance starting scope for a department-level workload.
By quotation
configuration and support scoped separately
- ArchitectureSingle appliance
- ScopeDepartment-level starting point
- ModelSelected after workload testing
- CapacityConfirmed from measured demand
- DeliveryWritten scope and acceptance criteria
Multi-team
Tower
البُرج
A multi-node starting scope for several teams or shared workloads.
By quotation
configuration and support scoped separately
- ArchitectureMulti-node deployment
- ScopeSeveral teams or shared workloads
- ModelsSelected after workload testing
- CapacityConfirmed from concurrency tests
- DeliveryWritten scope and acceptance criteria
Rack
الخزانة
A rack-scale starting scope for institution-wide workloads.
By quotation
custom infrastructure and support scope
- ArchitectureRack-scale GPU infrastructure
- ScopeInstitution-wide workloads
- ModelsSelected after workload testing
- FacilityReviewed before architecture
- DeliveryWritten scope and acceptance criteria
Full specification, fit and installation detail for each: Hosn Kernel, Hosn Tower, Hosn Rack, or the three compared side by side. Use the TCO planning guide to define a comparable cost boundary.
Before you scope a deployment
Sovereign AI Readiness Assessment
The assessment documents the workload, users, corpus, data boundary, integrations, facility constraints and acceptance method. Its output is a written scope that the institution's technical and procurement teams can review before a quotation is prepared.
By quotation
scoped to the institution, corpus and site
Request an assessment
04 · Use cases
What institutional teams can ask Hosn.
Approved document review
Workload pattern
مثاللخّص هذه المجموعة من الوثائق المعتمدة، واربط كل نتيجة بمصدرها داخل المجموعة.
ExampleSummarise this approved document set and link each result to its source within the set.
Arabic drafting support
Workload pattern
مثالأعد مسودة عربية من الملاحظات المعتمدة، واترك كل معلومة غير متوفرة محددة للمراجعة.
ExamplePrepare an Arabic draft from the approved notes and mark every unavailable fact for review.
Structured extraction
Workload pattern
مثالاستخرج الحقول المطلوبة من هذه النماذج المعتمدة، وأظهر الصفحة المصدر لكل قيمة.
ExampleExtract the required fields from these approved forms and show the source page for each value.
Internal knowledge search
Workload pattern
مثالاعثر على الإجراء المعتمد لهذا الموضوع، وقدّم الرابط والفقرة التي تدعم الإجابة.
ExampleFind the approved procedure for this topic and provide the link and paragraph supporting the answer.
Longer form on each of these workloads, and on archives, correspondence and records, on the use cases page and in the field notes cluster.
05 · Who Hosn is built for
Institutions whose duty of care does not end at the firewall.
A
Public institutions
Internal knowledge searchArabic drafting supportApproved document review
B
Regulated enterprises
Policy retrievalStructured extractionEvidence-linked summaries
C
Healthcare, research and professional services
Approved corpus searchBilingual knowledge workflowsHuman-reviewed drafting
06 · Deployment boundaries
Hosted AI or on-premise AI?
Hosted AI
Provider-managed infrastructure.
A network-connected service can simplify infrastructure management. Its data boundary, administration, logs and available models depend on the provider's documented terms.
حصن · Hosn
Customer-controlled infrastructure.
Hosn is deployed at the customer's site. Network isolation, local roles, logging and model selection are defined during discovery and recorded in the delivery scope.
Dimension
Hosted AI
Hosn
Deployment location
Provider-managed
Customer site
Network dependency
Service-dependent
Air-gap option
Administration
Shared with the provider
Customer-defined local roles
Model selection
Provider catalogue
Chosen during discovery
Logging and control
Service-dependent
Configured to local policy
Data boundary
Defined by provider terms
Defined in the delivery scope
Commercial model
Provider terms
Written quotation and contract
Deployment schedule
Provider-specific
Site and integration-specific
The right boundary depends on the workload, the institution's policy and the delivery requirements. Hosn documents the selected boundary before implementation.
Location
Muscat, Sultanate of Oman
Languages
Arabic and English
Deployment boundary
Air-gapped option
Questions
What institutions ask before the first meeting.
What is Hosn?
Hosn provides scope-driven on-premise AI planning and deployment for institutions in Oman. The model, infrastructure, integrations and acceptance criteria are confirmed through discovery.
Does Hosn require an internet connection?
An air-gapped deployment option can be scoped. The final network boundary, required routes and update procedure are documented for the delivery.
How is Hosn different from hosted AI?
Hosted AI runs in provider-managed infrastructure. Hosn is deployed at the customer's site, and its network and administrative boundary are documented for the delivery.
What does Hosn cost?
Pricing is by quotation because hardware, integration, deployment boundaries and support are scoped to the institution. The public site does not publish fixed prices.
Which AI models does Hosn run?
Selected open model families, including Gemma and Qwen, can be evaluated. The exact variant, runtime and context setting are confirmed after workload testing.
Does Hosn handle Arabic well?
Arabic and English requirements are tested against representative, approved material. The model and language acceptance criteria are confirmed before implementation.
How long does deployment take?
The deployment schedule depends on procurement, hardware delivery, site readiness, integration scope and acceptance testing. A delivery plan is included with the quotation.
What defines the software and model rights?
Software rights, upstream model terms, support and update responsibilities are stated in the written quotation and contract. No right is inferred from a public page.
Who is Hosn built for?
Hosn is intended for institutions in Oman that are evaluating an on-premise AI workload and need documented data, integration and acceptance boundaries.
Who builds and supports Hosn?
Hosn operates from Muscat, Oman. Public enquiries are handled through info@hosn.om and +968 9889 9100.