Custom AI models in Australia.
Ertas is an applied AI lab that builds custom AI models: models trained on your own work and shaped by your team. We work with organisations anywhere, and this page covers how it works if yours is in Australia. When your data has to stay in the country, the model runs in your own cloud account in an Australian region, or on your own servers.
What is a custom AI model?
A custom AI model is a language model trained on examples of your organisation's own work, so it learns your formats, terminology and decisions for a defined task. Before it takes any traffic, it is tested against a quality bar built from your own examples.
Most teams start with one task they already run at volume on a frontier model, such as document extraction, support triage or claims handling.
Can a custom AI model run in Australia?
Yes. If your data has to stay in Australia, run the model in your own cloud tenancy in an Australian region, or on-premise on your own servers. AWS, Microsoft Azure and Google Cloud each run cloud regions in New South Wales and Victoria, as of October 2026.
The region and instance type depend on the GPU capacity your account has there, and we size the model to fit it.
- Your cloud
- Your own AWS, Azure or Google Cloud account in an Australian region, under the controls you already have.
- On-premise
- Your own servers, inside your network.
- Ertas-managed
- We host, monitor and update the model, and your data stays private. Choose your own cloud or on-premise when the data must stay in Australia.
What does the Privacy Act mean for a custom model?
When personal information goes to a recipient overseas, Australian Privacy Principle 8 asks the organisation sending it to take reasonable steps so the recipient handles it in line with the Australian Privacy Principles, and section 16C can leave the sender accountable for the recipient's breaches. That is the OAIC's reading in its APP guidelines, chapter 8 (updated 3 October 2025). A model running in your own Australian region or on your own servers keeps its requests in the country.
Training raises its own questions. The OAIC's guidance on developing and training generative AI models (published 21 October 2024) says data being publicly available does not mean it can legally be used to train or fine-tune a model, and that using personal information collected for another purpose needs consent, or a use people would reasonably expect that relates to the original purpose.
So data preparation starts by removing personal information the task does not need, and your team approves what goes into the training set. This is general information, not legal advice.
Who builds and runs the model?
The team in Melbourne that runs the lab's research also builds and runs your model, so you work directly with the people doing the training. The published research shows how we measure models, and the proposed trust framework sets out evaluation, a safety floor and a model record.
Each engagement starts with a free audit of where your AI spend goes and which tasks repeat at volume. You get a written scope, and you decide whether to continue after every stage.
Common questions.
Do we own the custom model?
The model is trained on your work and shaped by your team, and it runs on Ertas by default. Ownership and handover of the model are available with a minimum engagement commitment.
Which clouds can run the model in Australia?
Your own AWS, Microsoft Azure or Google Cloud account, in an Australian region, or your own servers on-premise. We size the model to the GPU capacity you have there.
Where is the Ertas team based?
In Melbourne. Ertas AI is an applied AI lab that works with organisations anywhere, and the team that runs the research also builds and runs client models.
How do we start?
With a free audit of where your AI spend goes and which tasks repeat at volume. A proof of concept of three to four weeks follows, and pricing depends on the task, data volumes and deployment environment.