Key facts
| API surface | OpenAI-compatible /v1/chat/completions; keep your existing SDK |
| Arabic support | Multilingual chat and embedding models in the 30+ model catalogue |
| Deployment | Plugsky cloud with region selection, VPC, on-prem or air-gapped |
| Data residency | Sovereign deployment options for in-country processing |
| Auditability | Request and response logging, usage analytics and access controls |
| Retrieval | Embeddings and RAG are live for policy and service document search |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Audio, images, moderation, batch and fine-tuning are coming soon |
TL;DR
- Arabic-first retrieval works today with multilingual chat and embeddings.
- Residency is an architecture choice: region selection, VPC, on-prem or air-gapped.
- Every request can be logged for audit and accountability reviews.
- OpenAI compatibility keeps integration with existing systems simple.
- Pilot on public documents, then expand to controlled service workflows.
How it works, step by step
- Classify services by data sensitivity and decide what may be processed in a shared cloud.
- Choose the deployment target: in-region cloud for public content, private or air-gapped for restricted data.
- Build an Arabic and English evaluation set from real citizen questions with approved answers.
- Index service guides, policies and forms, and require answers to cite the indexed source.
- Define retention and access rules for logs, and restrict log access to authorized staff.
- Publish a pilot to one service channel and collect unanswered questions weekly.
- Set a review gate before any workflow can trigger a transactional action through a tool call.
Try it yourself
Open the AI data residency checklist →
Priority use cases for public services
The lowest-risk, highest-volume wins share one trait: they help staff or citizens find and understand information that already exists.
- Service guidance: answer questions about eligibility, documents and process from official guides.
- Internal knowledge: give caseworkers cited answers from circulars and procedures.
- Correspondence drafting: prepare Arabic and English replies for human approval.
- Content operations: summarize consultations and feedback without exposing personal data.
Transactional decisions stay with people; AI shortens the path to the decision.
Arabic-first retrieval
Arabic quality is a retrieval problem as much as a model problem. Use multilingual embeddings so Arabic questions match Arabic and mixed-language documents, normalize diacritics and spelling variants during ingestion, and keep an evaluation set of real citizen phrasing. Test both Modern Standard Arabic and the dialects your service actually receives.
Because the API is OpenAI-compatible, you can benchmark several multilingual models in the 30+ model catalogue without rewriting application code.
Residency and sovereignty in practice
Residency means controlling where processing and storage happen. Plugsky supports region selection for cloud deployments and private modes for stricter mandates: VPC hosting in your own tenancy, on-prem installation, and air-gapped environments with no outbound connectivity.
Document the data flow for each service, including logging and backups, and verify that no component calls outside the approved boundary. See the UAE residency architecture guide for a model you can adapt.
Auditability and accountability
An auditable AI service records what was asked, what was retrieved, which model answered and what was returned. Keep request IDs, timestamps, model names, retrieval references and outcomes under a defined retention schedule, and restrict who can read those logs.
Pair logs with a change process: when a model version changes, re-run the evaluation set and record the result before the change reaches production.
Honest comparison
| Capability | Plugsky | Consumer AI tools | Building in-house |
|---|---|---|---|
| Arabic quality | Multilingual chat and embedding models, benchmarkable in one API | Varies and not controllable | Depends on hosted model |
| Residency | Region selection, VPC, on-prem, air-gapped | Vendor-controlled regions | You control fully |
| Auditability | Request logging, usage analytics, scoped access | Limited or unavailable | You build logging |
| Deployment effort | Same code across cloud and private modes | Not deployable privately | Months of platform work |
| Pricing | Flat monthly self-serve plans; see live pricing | Per-seat subscriptions | GPU plus operations cost |
| Time to pilot | Days | Days, without residency control | Quarters |
Frequently asked questions
Can the service run fully inside our network?
Yes. Air-gapped deployment keeps prompts, documents and logs inside your environment with no outbound connectivity. The API shape stays OpenAI-compatible.
Is Arabic supported natively?
The catalogue includes multilingual chat and embedding models. Evaluate on real citizen questions, because dialect and domain coverage differ between models.
How do we prove the system is auditable?
Log request IDs, retrieved sources, model versions and outcomes under a retention policy, restrict log access, and keep evaluation records for each model change.
Do we need to expose personal data?
No. Redact or tokenize identifiers, retrieve only required fields, and keep authorization checks in your own services. For restricted data, use a private deployment.
Which endpoints are live for public services today?
Chat, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Audio, images, moderation, files, batch, assistants, responses and fine-tuning are coming soon.
How should we start?
Pilot with public documents and a small evaluation set on the free plan, then move to an in-region or private deployment once quality and controls are proven.