Australian businesses have been quietly routing sensitive client data through servers in California for years. Healthcare clinics, law firms, accounting practices: all feeding patient records, legal briefs, and financial statements into AI tools built and hosted under US jurisdiction. Most of them never stopped to ask where that data actually went. Their compliance teams are starting to ask now.
NorthStar LLM API is built specifically for this gap. It runs Gemma3 models on dedicated hardware in Adelaide, South Australia, served through Cloudflare’s Sydney and Melbourne edge nodes. No shared cloud. No US data centre. No prompts logged at any layer. The whole pitch is simple: private AI inference that stays inside Australian jurisdiction, full stop.
What makes it worth paying attention to is the timing. APRA CPS 230 requires Australian financial services entities to demonstrate third-party operational risk compliance by 1 July 2026. US-hosted AI providers cannot satisfy that requirement by design. The US CLOUD Act means data processed on their infrastructure is subject to US legal compulsion regardless of where it physically sits. NorthStar is purpose-built for exactly that problem.
Key Features of NorthStar LLM API
True Australian Data Sovereignty
Every inference request runs on dedicated hardware in South Australia. The request never touches a US server, never passes through US jurisdiction, and is never subject to US law. This is not a marketing claim with asterisks attached. The infrastructure is physically located in Adelaide, routed through Cloudflare’s Australian edge network. For healthcare providers, law firms, and financial services businesses, this is what compliance actually looks like in practice.
OpenAI-Compatible API with One Endpoint Change
If your team already has code talking to OpenAI, switching to NorthStar is a single line change. The API is OpenAI-compatible, which means existing Python scripts, curl commands, or application integrations work without rewriting anything. You swap the endpoint to https://api.northstarholdings.com.au, add your API key header, and you’re done. That’s a genuinely low barrier for any developer who wants to test it without committing to a full migration.
No Prompt Logging at Any Layer
NorthStar does not log or store prompts. Not at the application layer, not at the infrastructure layer. For regulated industries, this matters enormously. When a solicitor summarises a client file or a GP clinic processes patient notes, the content of those requests should not be retained on a third-party server. NorthStar treats this as a baseline requirement, not an optional privacy add-on.
Flat Monthly Pricing, No Per-Token Billing
Per-token billing from major providers creates unpredictable costs at scale. A team processing large volumes of documents in a month can face wildly different invoices each billing cycle. NorthStar uses flat monthly subscriptions, so costs are predictable regardless of usage volume. Two Gemma3 model tiers are available: the 4.3B parameter model for fast general inference, and the 12.2B parameter model for higher quality reasoning tasks.

A Real Use Case: Sophie at 9am on a Monday
Sophie Nguyen is a senior associate at a mid-size commercial law firm in Brisbane. Every Monday morning she works through the weekend’s client intake emails, pulling out key facts and drafting preliminary summaries before the 10am team meeting. She had been using a popular US-based AI assistant to speed up this process, until her firm’s risk partner flagged it in February: client communications were leaving Australian jurisdiction with every request.
At 9am on a Monday in early April, Sophie’s IT team had already switched her workflow to NorthStar LLM API. The integration was a single endpoint change in the firm’s internal summarisation tool. She submitted a 3,400-word client intake document to the API at 9:07am, asking for a structured summary with key obligations and risk flags identified. The response came back in under ten seconds, using the gemma3:12b model on the Pro tier.
The summary was clean and well-structured. More importantly, her risk partner could confirm exactly where the data had been processed: dedicated hardware in South Australia, under Australian law, with no retention of the prompt or output. The 10am meeting proceeded without anyone needing to caveat which client matters had been touched by an AI system operating outside Australian jurisdiction.
How the Request Actually Travels

Step 1: Your Application Sends a Request
Your app or script sends a POST request to the NorthStar endpoint with your API key and prompt. The format is identical to an OpenAI request, so existing tooling works without modification.
Step 2: Cloudflare Australian Edge Receives It
The request hits Cloudflare’s Sydney or Melbourne edge node. It stays inside Australia at this point. Nothing is routed offshore.
Step 3: NorthStar Gateway Processes and Forwards
The NorthStar Gateway authenticates the request and forwards it to the Gemma3 model running on dedicated Adelaide hardware. No logging occurs here.
Step 4: Response Returns to Your App
The model response travels back through the same Australian path. Your application receives it. The prompt and response are not stored anywhere in the chain.
Pricing
NorthStar offers flat monthly subscriptions with no lock-in contracts, giving developers a low-commitment entry point to test the integration. Paid plans are flat monthly subscriptions (all prices AUD):
- Developer: $29/month, gemma3:4b model
- Basic: $99/month, gemma3:4b model
- Pro: $199/month, gemma3:4b and gemma3:12b models
- Enterprise: $499/month, all models, priority support
No per-token billing surprises at any tier. You know what you’re paying before the month start
Who Should Take a Serious Look at This
If you’re building or maintaining AI-assisted tooling for an Australian healthcare provider, law firm, accounting practice, or financial services business, NorthStar LLM API deserves a proper evaluation. The compliance case is concrete, the integration overhead is minimal, and the pricing is predictable. With APRA CPS 230’s July 2026 deadline approaching, the window for quietly ignoring data residency questions is closing fast. Start at northstarholdings.com.au and see how quickly your existing code adapts.