/01What does a private AI deployment include?
A working deployment is more than a model. Every Sovata deployment combines a capable open-weight or licensable large language model, a retrieval layer that grounds answers in your approved documents and policies, access controls that decide who can see what, and logging so every interaction can be reviewed. The model supplies the language ability; the architecture around it supplies the accuracy, governance and audit trail that make it safe to use on real work.
- Architecture design matched to your security and governance requirements
- Private LLM selection and deployment — open-weight or licensable models, in your environment
- Knowledge preparation, ingestion and retrieval (RAG) over your own documents
- Access controls, boundaries and escalation paths agreed before go-live
- User onboarding, then ongoing monitoring, governance and improvement
/02Where can it run? Your three deployment options in New Zealand
On your own servers (on-premise)
The strongest sovereignty position: model, data and logs live in your server room or private rack. Suits organisations with existing infrastructure and strict residency or air-gap requirements — including deployments with no outbound internet access at all.
In a New Zealand data centre
Dedicated infrastructure hosted onshore, so data remains in New Zealand jurisdiction without you operating hardware. The common choice for organisations that need onshore residency and predictable performance without a capital purchase.
In your own cloud tenancy
A private deployment inside your organisation’s dedicated cloud environment in an onshore region. You keep the operational convenience of cloud while the model and data stay inside a tenancy you govern — not a vendor’s shared, multi-tenant platform.
/03How does this support NZ data sovereignty and the Privacy Act 2020?
Because you choose where the system runs, personal and regulated information is processed within a boundary you can point to on a map. That physical control is what public AI tools cannot offer: they can promise contractual protections, but usually cannot promise your data never leaves the country.
A private deployment is designed to support obligations under the Privacy Act 2020, the Health Information Privacy Code, and public-sector expectations around keeping citizen data onshore — and for Māori organisations, it keeps data under the governance of the people it belongs to. We define exactly what the system can access and do before go-live, and where a use case doesn’t fit, we say so.
/04What does private AI deployment cost?
Pricing has three stages rather than one number. AI Discovery is free. The AI Readiness Workshop is fixed-price. The deployment itself is scoped per organisation after the workshop, because the real cost drivers — data volume, governance complexity, infrastructure choice and integration scope — vary widely between organisations. You get a genuine range with the factors behind it before committing, and running costs after go-live are typically fixed and predictable rather than usage-metered.
/05How long does deployment take?
Most organisations go from readiness workshop to a working, governed system in 6–12 weeks, depending on data complexity and governance requirements. The four-step process — define the problem, set the boundaries, prepare the knowledge, deploy and operate — is documented step by step on our How It Works page.