Home / Services / AI and automation
Private AIYour AI. Hosted in Australia.
Custom models, fine-tuning and the systems around them, running on our own servers in an Australian data centre and answering only to you. For businesses that can't, or shouldn't, send their data to someone else's model.
Every question sent to a public AI service leaves your business.
A general AI knows a little about everything. Yours should know your business, and keep what it learns to itself.
- Client details pasted into public AI tools, with no record of where they end up.
- Per-use AI bills that grow with every person and every task.
- General models that don't know your products, terms or rules.
- Clients, auditors or regulators asking where their data is processed.
Your model, your data, your country
Custom models
A model built for one job in your business: reading your documents, sorting requests, spotting what needs attention, forecasting what comes next.
Fine-tuning
A proven open model trained further on your own material, so it speaks your language and follows your way of working.
Private hosting
Your model on our own servers in an Australian data centre. Your data isn't used to train models for anyone else.
A model only knows what it was trained on. We build everything around it.
The difference between an AI demo and AI that does real work is the system around the model: what it can look up, what it's allowed to do, and how its answers are checked.
Answers from your own documents
The model looks up your policies, contracts, manuals and records before it answers, and shows its sources. People only see what they're already allowed to see.
Wired into how you work
Reading from and writing to your systems, following your workflows step by step, and handing off to a person when it should.
Rules enforced by the system
Private details masked, actions it may never take, spending limits, and human approval for anything that matters.
Measured before it's trusted
A test set from your own real cases, run on every change, so a version that answers worse is caught first.
A predictable running cost, instead of a meter that keeps running.
Public AI services charge for every use, so the bill grows as your team relies on them more. A private model has a predictable running cost, and a model built for one job can often be smaller and cheaper to run than a general one.
Where else the money goesFor your IT team: how it's built and run
- Our own hardware, in AustraliaServers we own and operate, in a secure, access-controlled cabinet in a professional Australian data centre, with private network connections and on-site support.
- Private endpointsEach model is reachable only by the systems and people it serves, over encrypted connections, and is not exposed publicly.
- No outside AI providerPrompts, documents and outputs are processed on our servers rather than sent to an external AI service.
- Open models, chosen per taskProven open-weight models selected for the job and fine-tuned on your material.
- Retrieval with permissionsYour documents indexed in Australia and searched at question time, with each person's existing access applied and sources cited.
- Purpose-built harnessesTool access to your systems through narrow, typed actions, with step limits, timeouts, human hand-off and every call recorded.
- Evaluation suitesTest sets from your own cases, run on every model, prompt or retrieval change before release.
- Access control and auditWho used the model, when and for what is recorded.
- Encrypted and backed upTraining data, model versions and settings encrypted at rest and backed up, with restores tested.
- Versioned releasesEach model version is kept, so a new one can be rolled back if it performs worse.
Common questions
Where are the servers?
In a secure, access-controlled cabinet in a professional data centre in Australia. The servers are ours; your data is processed and stored there.
Is our data used to train anyone else's model?
No. Your material is used for your model only.
Which models do you use?
Proven open-weight models, chosen for the task and fine-tuned on your material where that helps.
Can it answer from our own documents?
Yes. Retrieval connects the model to your documents and records, applies each person's existing access, and shows the sources behind each answer.
Will it always be right?
No AI is right every time. That's why we test models against your own cases before release, show sources, and require a person's approval for anything that matters.
Tell us what's slowing you down.
Describe how things run today, or what worries you. We'll come back with a plain-English view: what we see, what we'd fix first, and roughly what it would involve. It's free, and yours to keep.