Sovereignty is a deployment decision, not a marketing label. The goal is simple: keep sensitive data, models, and inference inside your infrastructure while maintaining operational control and visibility.
1. Define Data Tiers
Start by classifying data with your security and risk teams. The tiers below are an internal planning aid — use your own policy definitions.
| Tier | Examples | Suggested Deployment |
|---|---|---|
| Public | Public documentation, marketing | Private cloud or public API (if allowed) |
| Internal | Employee FAQs, internal SOPs | Private cloud |
| Restricted | Customer data, financial reports | Private cloud with strict controls |
| Highly Restricted | Critical infrastructure, national security | Air‑gapped on‑prem |
2. Pick a Deployment Model
Choose a model that fits your tiering policy. Atlas supports private cloud deployments and air‑gapped environments, with routing and isolation built into the runtime.
Deployment rules of thumb
- Keep restricted data inside your VPC or on‑prem boundary.
- Disable external routing by default; enable only when necessary.
- Use local activity journaling for operational visibility.
3. Deployment Example
This example shows a private‑cloud rollout that keeps sensitive data inside a dedicated VPC while preserving operational control.
- Provision a private VPC and isolate subnets for inference nodes.
- Deploy Atlas Runtime and Core into the VPC using your IaC templates.
- Enable local activity journaling and verify routing defaults.
deployment:
mode: private-cloud
network: vpc-isolated
routing:
external: disabled
telemetry:
journaling: enabled4. Operational Controls
Define the controls that keep sovereignty intact: access boundaries, model routing defaults, and retention policies aligned with your internal security standards.
- Restrict who can change routing or model versions.
- Store logs locally and enforce retention windows.
- Review access regularly and rotate credentials on a cadence.
5. Runbook & Ownership
Sovereignty is a continuous practice. Define ownership, escalation, and change management so production deployments stay stable as models and data evolve.
- Document model upgrades and rollback plans.
- Align retention policies with internal security policies.
- Review telemetry and routing decisions regularly.