ArabyAI cloud
We run it for you on our cloud. Nothing to install.
For companies that want to start fastSecure AI for government and enterprise.
Your data stays where you decide: our cloud, your national cloud, your servers, or fully offline. And your security office can check every step.
Your data stays where you decide.And you can check it.
The first question every organization asks. There are four ways to run it, and you choose.
Your national or government cloud
Data stays in your country, on the cloud your entity already uses.
For government and regulated sectorsOn your own servers
We install it inside your data center, run by your team.
For organizations with their own data centersFully offline
No internet connection at any point, for the most sensitive work.
For secure and sovereign environments
Edge gateway
GPU cluster
Key vault
Audit log
External AI blocked
Records and vectors
UAE Pass · ICPAllowed
Edge gateway01 / 05
The one way in. TLS 1.3 at the edge, mutual TLS between every internal service, plus WAF and rate limits.
GPU cluster02 / 05
Open-weight models on your own GPUs, or approved models through the Federal AI Gateway. Never a public AI service.
Key vault03 / 05
Records and vectors are encrypted at rest with AES-256. The keys live in your HSM, not ours.
Audit log04 / 05
Every agent action is written to an append-only log you own, ready to export to your SIEM.
Egress gate05 / 05
Outbound traffic is denied by default. Only allow-listed government endpoints, such as the Federal AI Gateway, UAE Pass and ICP, are reachable.
Your data neverleaves the building.
The whole agent stack runs inside your perimeter, on your premises or in a UAE sovereign cloud. Models run on your GPUs or through the Federal AI Gateway, keys stay in your vault, and every action is logged. Residency is built in, not promised in a contract.
0
calls to external AI
1
way out, to services you approve
Outbound traffic is denied by default. Only the government endpoints you allow-list are reachable.
Six practicesyour security office can approve.
What makes an agent safe to run inside your institution, on every project.
- (01)
Least privilege
The agent acts as the user it serves. What the user cannot see or do, the agent cannot either.
- (02)
Approved tools only
Every action is a tool you approved, with a scope and validated inputs.
- (03)
Guardrails, tested
Prompt-injection defences, personal-data and refusal rules, tested on your cases before go-live.
- (04)
Human handover
Anything out of scope goes to a person with the full context attached.
- (05)
Full audit
Every request, model call, tool call and reply is logged and replayable, inside your environment.
- (06)
Evaluation before release
A test set from your real conversations runs before every release, against pass marks you set.
Built for government.Compliant by design.
Our agents follow the AI Office and TDRA guideline for the Federal AI Gateway and the national guide for agentic government services. Onboarding is a form to complete, not a rebuild.
PeopleCitizens and staff
Citizens
Staff
Duty officer
Agentic applicationInside your environment. Built and run by Araby.ai.
Agent orchestrationPlans, acts, follows up
IntegrationsUAE Pass, ICP, your systems
Knowledge and vectorsYour documents, cited
Memory and historyKept in your environment
GuardrailsRedaction and policy checks
Key vaultOne Gateway key per entity
Audit logEvery action, append-only
Human approvalsConsent and sign-off
Federal AI GatewayThe single governed route to models
Entity API key
Authentication
Sensitive-data redaction
Prompt protection
Quotas and rate limits
Routing and failover
Metering
Approved modelsConsumed as a service
Sovereign-hostedApproved models
Managed commercialApproved models
01 / 04
One governed route to models.
- Every model call goes through the Gateway endpoint. Nothing reaches a model provider directly.
- A separate API key for each entity, never shared between clients.
- Keys kept in a vault and rotated whenever the Gateway reissues them.
- Use case, models, expected TPM and data classification declared in the onboarding form.
- Changes to scope or usage raised with the Gateway team first.
SourceFEDAI Gateway Onboarding Guideline v1.5, AI Office with TDRA
02 / 04
Designed the way the national guide asks.
- Uses data the government already holds, so people are never asked twice.
- Confirms the request and explains the steps before it starts.
- Asks for consent before any action is taken.
- Hands over to an employee with the full context when needed.
- Reports a clear outcome and lets people correct or object.
SourceGuide to designing government experiences with agentic AI, National Committee for Agentic AI
03 / 04
Your data, your classification.
- Your entity sets the data classification. We design the controls to match it.
- Conversation history, agent memory, RAG content and vector databases stay in your environment.
- Sensitive fields are redacted before each Gateway call, on top of the Gateway's own redaction.
- Outbound traffic is denied by default. Only the Gateway and approved government endpoints are reachable.
SourceFEDAI guideline, service and data responsibility · UAE Information Assurance Regulation
04 / 04
People stay accountable.
- Every agent action is written to an audit log your team can review.
- Consequential steps need a person's approval.
- Model output is never the sole basis for a decision.
- We own incident response at the application layer and escalate Gateway issues to the Gateway team.
SourceUAE Charter for the Development and Use of AI: human oversight, governance and accountability
Aligned withFEDAI Gateway Onboarding Guideline v1.5National agentic service design guideUAE AI CharterUAE Information Assurance Regulation
Who sees what.And how you check it.
We describe what we do plainly: where data lives, who can reach it, and what gets logged.
Access by role
Each person sees only what their role allows, and our team sees nothing without your approval.
A log of every request
Every request and action is recorded, so your security office can review it.
Your permissions, unchanged
When we connect to your systems, we keep each user's existing permissions.
The security document for your security office
A short file that answers the first approval questions: where data lives, who can reach it, and what gets logged.
We name a certification, standard or law on this site only once we hold or meet it, with proof ready.
Securityquestions.
The questions your security office will ask first, answered plainly.
Where you choose, from the four ways to run it: our cloud, your national cloud, your servers, or fully offline.
Yes. In a fully offline deployment everything runs inside your network. [confirm with the tech team]
Only the people you allow. Access is by role, our team sees nothing without your approval, and every request is logged.
Approved models only, chosen with your security office. Where the Federal AI Gateway applies, every model call goes through it.
Bring us one process. We will show you the AI working.
A one-hour meeting with our team, on a real example from your organization.