Sovereign AI, hosted in Australia

AI that holds up when someone challenges it

We build evidence-grade AI for Australian firms whose work gets audited, contested or underwritten. Models hosted in Australia. Every output traced to its source, logged, and tested before it ships.

Client data never leaves Australia. No prompts or documents are sent to offshore AI providers. Prefer to talk? 0416 883 436

Astronaut guarding a vault shaped like Australia, representing data kept onshore

Where your data lives

Data sovereignty is the whole design, not a clause in the terms. Your documents, the index built from them, the model that reads them and the audit log all sit inside Australia. Nothing crosses that line.

Hosted in Australia: the model runs in an Australian cloud region in Sydney, in a private environment we manage for your firm. On premise: for firms that want it, the same stack runs as a self-hosted LLM inside your own infrastructure. We agree which one before the build starts, and it is written into the engagement.

Inside Australia
  1. 01Your documentsClient files, briefs, ledgers, loan packs
  2. 02Private evidence indexSearchable, access-controlled, yours
  3. 03Model hosted in AustraliaAnswers only from what was retrieved
  4. 04Cited answer + audit logEvery claim linked to its source page
Offshore AI providersNo connection. No prompts, files or outputs sent.

What evidence-grade means

Four guarantees on every build. They are what make private AI auditable, and what let you defend its work.

Hosted in Australia.

Models and data stay onshore.

Traceable.

Every output links back to the source document and page it came from.

Logged.

A complete audit trail of what was asked, what was retrieved and what was returned.

Tested.

Evaluated against known answers before it goes near a client file.

Who this is for

Partners and principals at expert firms of ten to fifty people who want AI and cannot sign off on sending a client's financials, a brief of evidence or a credit file to an offshore model. Our clients are not one sector. They share a shape: scarce senior staff, confidential client data, personal liability, and work that someone hostile will examine line by line.

A tax and CFO advisory practice

Tax, SMSF, restructures and outsourced CFO work for private clients.

Answers toThe ATO

A construction claims practice

Forensic delay and quantum analysis, and expert witness reports.

Answers toA court or arbitrator

A commercial property debt adviser

Arranging property finance across a panel of more than 100 lenders.

Answers toA credit committee

What we build

Grouped by the job it does, so you can find your own problem. Every one ships with the four guarantees.

Turning documents into answers

  • RAG agentsAnswer questions from your own files, with each claim cited to the page it came from.
  • Evidence-linked knowledge searchSearch years of matters, briefs or loan files and land on the exact paragraph, not a summary of it.

Checking and reconciling

  • Reconciliation at scale →Match thousands of transactions, invoices or schedules and surface only the exceptions.
  • Spreadsheet agentsRead, check and update the workbooks your firm already runs on, with every change recorded.

Producing the deliverable

  • Report generationDraft the first version of a report in your house format, with sources attached for the reviewer.
  • Capital statement generatorAssemble capital and funding statements from the source records, ready for sign-off.

Modelling a decision

  • Estimation engineBuild estimates from historical data and show which records drove each number.
  • Feasibility assessmentRun development feasibility with every assumption stated and traceable.
  • Debt dashboard analyticsTrack a debt book across lenders and facilities from one audited data set.

Proof

50 → 110

clients handled by one bookkeeping practice, with the same team.

20h → 2h

monthly bank reconciliation for a Brisbane accounting firm.

The same method runs in tax advisory, construction claims and property finance. It travels on process, not industry. See the accounting results.

How every build is verified

  1. An evaluation set from your own work. Before launch we build a set of real questions from past matters, with answers your senior people have already agreed are right.
  2. Your expert signs off accuracy. The system is scored against that set, and a partner at your firm, not us, decides whether it is good enough to use.
  3. It says "I don't know". When the evidence is thin, it stops and escalates to a person rather than filling the gap. That is a feature.
  4. Re-tested on every change. A new model version or new documents means the evaluation runs again before anyone relies on it. How we evaluate models.

Questions partners ask us

Where does the model actually run?

In one of two places, and we tell you which before we start. Hosted builds run in an Australian cloud region in Sydney, in an environment we manage for you. Builds on your infrastructure run entirely inside your own environment, on premise or in your own Australian cloud tenancy. In both cases no prompt, document or output is sent to an offshore AI provider.

How do I know it isn't making things up?

Every claim in an output points back to the source document and the page it came from. A reviewer clicks the citation and checks it in seconds rather than trusting it. If the system cannot find support in your documents, it says so instead of guessing.

What happens when it gets something wrong?

When confidence is low, the system flags the answer and routes it to a person instead of presenting it as settled. Nothing reaches a client without a qualified reviewer from your firm signing it off. We expect errors; the design goal is that none go undetected.

Can I show this to my regulator, my insurer or the other side?

Yes. The audit log records who asked what, which documents were retrieved, which model version answered and what was returned, with timestamps. Retention is set to match your record-keeping obligations, and the log exports in a readable format for an auditor, an insurer or a tribunal.

We are too small for this.

Firms of ten to fifty people are who we build for, not enterprises. Projects typically run $20,000 to $50,000, scoped to one workflow that already costs your senior people hours every week.

How long until it works?

A typical build goes from first conversation to a working pilot on your own files in about three months. You see it answering real questions, with citations, before you commit to rolling it out.

Book a technical review

A conversation about architecture, not a sales call. Bring the workflow you want to automate and the question your regulator, insurer or client would ask about it.

  1. 01We reply within one business day.
  2. 02We map where your data would sit and how outputs are checked.
  3. 03You get a written scope and price, or a straight answer that it is not a fit.

Or call 0416 883 436

A technical person replies within one business day, not a salesperson.