Government & semi-government
Private AI for government entities in Saudi Arabia and the GCC
Private AI gives government and semi-government entities in the GCC AI assistants and agents that run inside their own data centres or an in-country sovereign cloud, so official records stay under the entity's control. Seekers AI builds Arabic knowledge assistants for staff, citizen-service agents, document routing, and correspondence drafting. Each system follows the entity's data classification, access rules, and approval chain.
Definition
What is private AI for government?
Private AI for government means AI models, search, and agents deployed on infrastructure the entity controls or has approved, connected to its own records. No data goes to a public AI service. The entity sets who can use each assistant, what it can read under data classification rules, and which steps need a named official's approval.
Why don't public AI tools fit government entities in the GCC?
Government entities hold some of the most sensitive data in the country: citizen and resident records, internal decisions, contracts, and correspondence between ministries.
Public AI tool
Your organisation
Outside your control, often abroad
Data leaves
Private AI with Seekers
Your environment
Nothing leaves
Read moreShow less
Many entities must keep this data in-country, and national data classification rules decide which records can go where. A public AI tool that sends prompts to servers abroad is usually ruled out before anyone tests it.
In Saudi Arabia, the National Data Management Office (NDMO) data governance standards and the Personal Data Protection Law (PDPL), enforced by SDAIA since September 2024, set rules for how data is classified, handled, and transferred. National Cybersecurity Authority (NCA) controls apply to government systems. Other GCC states have their own data protection laws and cloud policies, and most entities add internal rules.
The work is also in Arabic. Circulars, decisions, memos, and citizen requests are written in formal Arabic, often with English attachments, handwritten notes, or scanned stamps. Assistants built mainly for English miss terms, misread names, and produce Arabic that staff would not send out. Many entities also run older correspondence and archive systems that public tools cannot connect to.
What does private AI look like inside a government entity?
We deploy Seekers Core, our private AI stack.
◇ Governance wraps every step
- 01ConnectDocuments, ERP, CRM, email
- 02UnderstandArabic + English search, cited
- 03ReasonPrivate models plan the task
- 04ActDraft, file, route, update
Read moreShow less
Connect links it to the entity's correspondence system, document archive, intranet, email, and service platforms. Understand indexes those records in Arabic and English and returns answers with the source document and paragraph.
Reason uses a private language model to plan tasks, such as reading an incoming letter and working out which department owns it. Act lets an agent carry out the task: route the letter, draft the reply, or update the request.
Governance covers every stage: access follows roles and classification levels, every action is logged, sensitive steps need human approval, and we monitor accuracy and drift after go-live.
Which deployment fits a government entity: on-premise, private cloud, or sovereign GCC cloud?
The choice usually follows the classification of the data involved and the infrastructure the entity has already approved.
- On-premiseData lives: Your own data centre
Best when: Strictest data rules, or air-gapped systems
- Private cloudData lives: Your dedicated cloud tenancy
Best when: You already run on a private cloud
- Sovereign GCC cloudData lives: An approved in-country cloud region
Best when: You want speed without buying hardware
Read moreShow less
Many entities use more than one. A citizen-service agent that answers from published service guides can run in a sovereign cloud, while an assistant over internal correspondence stays on-premise.
- On-premise: everything runs in the entity's own data centre, and the models need no external connection. This fits restricted data and entities with existing server capacity.
- Private cloud: an isolated environment in a government or shared-services cloud the entity already uses. This fits entities whose IT is run centrally by a shared-services body.
- Sovereign GCC cloud: an in-country cloud region run under national rules. This fits lower-classification workloads, such as public service information.
How does a government engagement with Seekers AI start?
We start with the Private AI Launchpad.
2–3 weeksUse cases, architecture, costed roadmap
4–8 weeksOne use case live on your systems
MonthlyOperate, measure, add use cases
- Readiness Sprint · 2–3 weeksUse cases, architecture, costed roadmap
- Pilot · 4–8 weeksOne use case live on your systems
- Run & Scale · MonthlyOperate, measure, add use cases
Read moreShow less
The Readiness Sprint takes 2 to 3 weeks. We work with the business owners, IT, information security, and the entity's data management office to list candidate use cases, check data classification and system access, and pick one to build first.
The sprint ends with an architecture, a risk review, and a costed roadmap that your internal committees can review. The Pilot (4 to 8 weeks) puts one use case into production on approved infrastructure with real staff. Run & Scale is a monthly service in which we operate the system, report on accuracy, and add use cases.
Which government use cases come first?
- 01
Arabic knowledge assistant for staff
Reads
- Document archive or content management system
- intranet
- legal library
- +2
Does
Answers staff questions about regulations, circulars, internal procedures, and past decisions in formal Arabic or English.
Governed by
Answers draw only on documents the user's role and clearance can already open.
Read moreShow less
- What it does
- Answers staff questions about regulations, circulars, internal procedures, and past decisions in formal Arabic or English. Each answer cites the document, its date, and the paragraph, and says when a newer circular has replaced an older one.
- Systems it connects to
- Document archive or content management system, intranet, legal library, shared drives, and the entity's directory service.
- Governance
- Answers draw only on documents the user's role and clearance can already open. Every query is logged. The legal or policy department controls which sources are indexed.
- 02
Citizen-service agent
Reads
- Service portal
- mobile app
- request or case management system
- +2
Does
Answers citizen and resident questions on the website, app, or WhatsApp in Arabic and English, using published service guides.
Governed by
The agent uses only approved, published content plus read access to request status.
Read moreShow less
- What it does
- Answers citizen and resident questions on the website, app, or WhatsApp in Arabic and English, using published service guides. It checks request status, explains which documents a service needs, and hands the case to an employee when it needs a decision.
- Systems it connects to
- Service portal and mobile app, request or case management system, contact centre platform, and approved messaging channels.
- Governance
- The agent uses only approved, published content plus read access to request status. It cannot approve, reject, or charge fees. Complaints and sensitive cases go to staff, and conversations are kept under the entity's records policy.
- 03
Document routing
Reads
- Correspondence management system
- scanning
- +2
Does
Reads incoming letters, emails, and scanned documents, identifies the sender, subject, and required action, and routes each item to the right department with a short Arabic summary and a suggested deadline.
Governed by
Routing rules come from the entity.
Read moreShow less
- What it does
- Reads incoming letters, emails, and scanned documents, identifies the sender, subject, and required action, and routes each item to the right department with a short Arabic summary and a suggested deadline.
- Systems it connects to
- Correspondence management system, email, scanning and document capture, and the organisation chart or workflow engine.
- Governance
- Routing rules come from the entity. Uncertain or classified items go to the registry team. Every routing decision is recorded with its reason.
- 04
Correspondence drafting
Reads
- Correspondence system
- document archive
- template library
- +1
Does
Drafts replies, memos, and letters in formal Arabic using the entity's templates and style, with facts taken from the incoming letter, the case file, and related past correspondence.
Governed by
Nothing is sent without review.
Read moreShow less
- What it does
- Drafts replies, memos, and letters in formal Arabic using the entity's templates and style, with facts taken from the incoming letter, the case file, and related past correspondence.
- Systems it connects to
- Correspondence system, document archive, template library, and email.
- Governance
- Nothing is sent without review. The responsible employee edits and approves each draft, and signing authority stays with the official who holds it. The audit log keeps the draft, the edits, and the approver.
Questions government entities ask about private AI
Can private AI run without an internet connection?
Yes. In an on-premise deployment, the models, search index, and agents run inside the entity's network without calling external services. Model and software updates go through the entity's own change process.
How does the system handle classified documents?
During the Readiness Sprint we map the entity's data classification levels to access rules. Documents keep their classification once indexed, users only get answers from documents they are cleared to read, and higher-classification data can sit in a separate deployment. The entity's information security team reviews this design before anything goes live.
How well does the AI handle formal government Arabic?
We test on the entity's own documents rather than translated samples. During the Readiness Sprint and Pilot, staff review answers and drafts for accuracy, terminology, and tone, and we adjust retrieval, prompts, and templates from their feedback. We keep reporting output quality during Run & Scale.
Does private AI meet our national data regulations?
We design each deployment around the requirements the entity gives us, including the PDPL, NDMO standards, NCA controls, and in-country data rules. Whether it complies is for the entity and the relevant authorities to decide. We provide the architecture, data flow, and controls documentation for that review.
Start with a Readiness Sprint for your entity.
In 2 to 3 weeks, we review your documents, systems, and classification rules with your business, IT, and information security teams. You leave with a use-case shortlist, a private AI architecture, a risk review, and a costed roadmap, whether or not you build with us.