Is a private model as good as ChatGPT or Claude?

For bounded business tasks, current open-weight models are close enough that most teams cannot tell. For frontier-level general reasoning, the hybrid setup routes non-sensitive work to frontier models under no-training agreements.

Where does our data live?

In your own cloud account in an NZ or AU region, or on your own hardware. We build in your environment, not ours.

Does cloud hosting really keep us compliant?

It depends on your obligations. Some client agreements are satisfied by NZ data residency and zero retention. The strictest require on-premise. We tell you honestly which tier your obligations actually require at discovery.

Private AI

Your own AI. Your own infrastructure. Nobody else's terms.

Previewlabs builds private AI systems for NZ and AU firms that cannot send client data to public AI services. We deploy open-weight models on infrastructure you control, in your own cloud account or on your own hardware, with NZ or AU data residency. Your data never trains anyone else's model.

Already running. Deployed for two Australian companies in the first quarter of 2026. Both under NDA. Built by a team that includes two machine learning engineers and a data security specialist with more than two decades in the field.

The problem, plainly.

Your team is already using AI. The question is whose. Every prompt pasted into a public chatbot is client data leaving your control. For a law firm, that is privileged material. For a clinic, that is health information. The answer is not banning AI. It is owning it.

Three ways to own it

Private cloud. A dedicated, isolated AI environment in your own cloud account. NZ or AU data residency. Zero retention. Live fastest.

Your own model. An open-weight model you own outright, tuned on your documents and language where it earns its keep, served on dedicated GPUs. No per-seat AI fees. No paywall between you and your own system.

Fully on-premise. Your hardware, your building, no outside connection. For the firms whose clients demand it.

Full agent systems, privately

Everything we build elsewhere runs here too. Document analysis and drafting, intake, matter and client workflows, internal assistants, full agent systems. Private does not mean limited.

The hybrid rule

Confidential work runs on your private model. General intelligence, the everyday drafting and research that touches nothing sensitive, can still use frontier models under strict no-training agreements. You get the best of both without betting your compliance on either.

Built for

Law firms and conveyancers. Healthcare and allied health. Accounting and financial advisers. Any firm where "we sent it to a US chatbot" is not an acceptable sentence.

How it works.

01 · Discover

We map how work really happens.

02 · Build

We deploy into your existing stack.

03 · Stick

We stay three months after launch.

Pricing is scoped after a free discovery call.

Questions, answered.

For bounded business tasks, current open-weight models are close enough that most teams cannot tell. For frontier-level general reasoning, the hybrid setup routes non-sensitive work to frontier models under no-training agreements.

Every engagement starts with a conversation.

Book a free 30-minute discovery call. We will look at your operations, identify what is worth automating, and tell you honestly whether we are a fit.