AI for national security — zero external dependency

Sovereign AI for national security

Defence and national-security organizations have the strictest AI requirements in existence: data that cannot leave the country, models that must run without external dependencies, and systems that keep working when internet access is restricted. That is the definition of sovereign AI — and it is the tier Plugsky built its on-prem and air-gapped offerings for.

Deployment tiers for security organizations

Tier	Isolation	Typical use
On-prem LLM	Everything inside your perimeter	Classified and air-gapped workloads
Private endpoint (VPC)	Dedicated, network-isolated	Operational analysis
In-region API	Regional data residency	Unclassified support workloads

Open-weight models, no vendor lock-in

Open-weight models (Nemotron, Llama, Qwen, Mistral) can be vetted, re-hosted, and run indefinitely — no proprietary runtime, no foreign API dependency, no export-control surprises. Plugsky delivers the models you choose, on hardware you control.

Security-relevant features

• Audit logs — every request is logged for SIEM export.
• BYOK — customer-managed keys.
• No data egress — on-prem tier keeps everything inside.
• Right-to-audit — available in enterprise agreements.

Defence AI FAQ

Can models run fully disconnected?

Yes — the on-prem tier runs without internet. See on-prem LLM for requirements.

Is the API compatible with existing systems?

Yes — OpenAI-compatible, so integration with existing tooling is a base-URL change.

Do you offer classified-level controls?

We provide the architecture (on-prem, air-gapped, BYOK, audit); specific accreditation is done with your security team. See security.