Sovereign AI Box

Description

A self contained AI server, in your office, owned by you. Canadian built, Canadian deployed, Canadian supported. No cloud. No telemetry. No data leaving the building. Designed for organizations with data residency requirements, sensitive operations, or a strong preference for not renting their AI from a US hyperscaler.

What is in the box:

Hardware. Workstation class server with discrete GPU sized to your inference load. Ryzen or Epyc compute, 64 to 256 GB RAM, NVMe storage with mirror. Quiet enough for an office, robust enough for production.

Software stack. Linux base, Ollama for local inference, LiteLLM proxy, your choice of open models, agent runtime, optional Claude Code with API egress firewalled to Anthropic only.

Network isolation. VLAN separated, default deny outbound. The only egress is the destinations you explicitly authorize. Nothing leaks to vendor telemetry, model providers, or third party monitoring.

Monitoring you control. Prometheus and Grafana installed locally, dashboards visible to your operators only. Optional encrypted tunnel for our remote support, on a switch you can flip off.

Documented operation. Every system has a runbook. Backup, restore, patching, model updates, capacity planning. Your team can operate the box without us if they choose to.

Canadian support. Vancouver based engineer answering the phone. On site visits within driving range of the Lower Mainland. Air shipped support hardware anywhere in Canada.

Pricing: setup $8K to $15K (one time, hardware and deployment), plus optional monthly support $1K to $2K (patching, monitoring, model updates, incident response, cancel any time).