Skip to content
Cyphonica

Sovereign AI

Local & NZ-Hosted AI Deployments

Your AI. Your hardware (or ours, in NZ). Your data never leaves the country.

A working LLM-backed assistant or agent running entirely on infrastructure you control, with no offshore data flow and no per-token surprises.

Who this is for

  • NZ businesses that want AI productivity without sending company data to OpenAI, Anthropic or Google
  • Regulated sectors (health, legal, finance, government-adjacent) where data sovereignty is non-negotiable
  • Teams who have hit the wall with public-cloud AI on cost, privacy or latency
  • Operators who want their own internal "Copilot-style" assistant on their own documents, not a third-party SaaS

What we actually do

  1. Use-case scoping - what should the model actually do, and is local the right answer
  2. Model selection from open-weight families (Qwen, Llama, Mistral, etc.) sized to your hardware
  3. Deployment on your servers, your VMs, or our NZ-hosted infrastructure - your choice
  4. Optional retrieval pipeline over your own documents (private RAG, no data leaves the box)
  5. Agent harness with tool-calling, audit logging and rate limiting (the same loomctl architecture we run internally)
  6. Bearer-token auth, role separation, full request/response trace storage for compliance
  7. Integration into VS Code, internal tooling or a private web UI
  8. Handover documentation and operator training

What you get

  • Working local or NZ-hosted AI service inside your environment
  • Configuration in source control, reproducible and yours to keep
  • Operations runbook covering model updates, monitoring and incident response
  • Optional ongoing support and quarterly model refresh

Pricing

Proof-of-concept from $4,000 + GST. Production deployment scoped on use case and hardware. Ongoing NZ hosting priced separately.

Timeline

PoC inside two weeks. Production deployment typically four to eight weeks depending on integration depth.

How we get access

Access to the target infrastructure, or a clear hardware spec if we are sizing it for you. No data egress required.

Common questions

Why local AI instead of just using ChatGPT or Copilot?
Three reasons: data sovereignty (your IP, client data and internal docs never leave your environment), cost predictability (fixed hardware cost vs per-token billing that scales with usage), and control (you choose the model, you choose when it updates, you keep working if the offshore provider has an outage or changes their terms).
Will it be as good as GPT-4 or Claude?
For specific, well-scoped tasks against your own data, modern open-weight models are genuinely competitive. For everything-to-everyone general chat, no. The honest answer comes out of the scoping conversation - we will tell you when local is the right call and when it isn't.
Where does "NZ-hosted" actually live?
On your hardware in your office, on a VM in a NZ data centre under your account, or on our NZ-based infrastructure under contract. All three options keep data inside New Zealand. We document exactly where everything sits.

Ready to scope this?

Send us your details. We come back with a fixed price and a start date.

Get a quote