Private AI agents for Saudi and GCC enterprises
Where a prompt is allowed to go depends on what it touches — not on which model happens to be fastest or cheapest. Private AI means you decide which data crosses the boundary, which models answer which questions, and where inference runs.
Customer-controlled inference and execution boundaries.
What private AI means
Private AI starts with a documented processing boundary: the model endpoint, control plane, command execution, evidence storage, retention and support access. Agree these together. Command-risk review is distinct from document classification or DLP; establish any data labels and permitted destinations during design.
Example data-handling policy to agree
| Confidential / Restricted | Internal cluster only |
| Internal | Internal cluster or an approved SaaS model |
| Public | An approved external model |
Boundary decisions to verify
- External inference
- Internet access
- Telemetry export
- Prompt retention
- Local models (via the gateway)
- Air-gapped operation
Bring your own model
Agree a supported model endpoint and test it with the selected deployment. Model location, credential handling, inference costs and allowed data must be recorded in the technical scope.
Bring your own model — OpenAI-compatible endpoints, Ollama, Anthropic, or Cloudflare AI Gateway / Workers AI — with keys envelope-encrypted at rest.
Who is signed in
Customer-hosted installations support OIDC with PKCE. Confirm workspace sign-in and SSO separately for the hosted deployment.
Customer-hosted installations support an identity provider through OIDC with PKCE. Hosted workspace SSO must be confirmed for the agreed deployment.
Private AI vs public AI agents
When comparing a hosted service with a private deployment, ask where prompts are processed, who can access them and which retention rules apply. Review NCA ECC, PDPL and SAMA CSF with your security and privacy teams to identify the requirements that apply to your organisation. A private deployment alone does not establish compliance.
Start with a six-week pilot
Plan a six-week pilot around one or two use cases from your backlog. Agree the deployment boundary, read access, evidence to review and success criteria before starting. See pricing for the indicative pilot cost and what happens next.
PricingStart free trialEvaluate your private AI deployment
Choose a deployment guide, then review the controls with the team that will operate it.
Private AI agents for Saudi and GCC enterprises