FAQ
Questions, answered straight
Including the ones with answers that do not suit us.
The basics
- What is an on-premise AI system?
- A dedicated machine that lives on your network and runs capable open AI models locally. Your team uses it through a browser, much like any AI chat tool. The difference is that the request never leaves your building — there is no external service to send it to.
- How is this different from using ChatGPT?
- Architecturally rather than cosmetically. With a cloud service, your text is transmitted to a third party, usually processed overseas, and retained for some period under terms they set and can change. With an on-premise system, the model runs on hardware you own — nothing is transmitted, so none of those questions arise.
- Do we need internet access for it to work?
- No. Once installed it runs entirely offline. It needs to be on your local network so staff can reach it, but it does not need an internet connection to function. That is genuinely useful at remote sites and during outages.
- What does the team actually see?
- A chat page in their browser. Type a question, paste or upload a document, get an answer. Anyone who has used an AI chat tool will need essentially no training on the interface — the training we provide is about what to trust and what to check.
Capability
- Is it as good as ChatGPT or Claude?
- For summarisation, drafting from your own material, extraction and Q&A over your documents — the work that fills most professional days — the open models available now are strong, and the gap is smaller than most people expect. For frontier reasoning on genuinely novel problems, the large commercial models still lead. We will tell you which of your use cases fall on which side before you spend anything.
- Which models does it run?
- Open models — the strongest available that fit the hardware, chosen against your actual workload. The ecosystem moves quickly and better releases arrive regularly, so we select at build time rather than committing to a name that will be outdated in six months. New models can generally be installed on the same hardware later.
- How fast is it?
- It depends heavily on the model, and any vendor quoting one number without naming the model is not being straight with you. Modern mixture-of-experts models run quickly because only a fraction of the model activates per token. Large traditional dense models are considerably slower on the same hardware — sometimes slow enough to be irritating for interactive use. We will show you real speeds on the models you would actually use.
- Can it search our existing documents?
- It can be configured to answer questions across a document library you point it at. How cleanly that works depends on where your documents live and what that system exposes. We establish this during scoping rather than promising integration and later discovering the API does not exist.
- Will it make things up?
- Sometimes, like every AI system including the largest commercial ones. Output must be reviewed by the person responsible for it. We build this into the handover training deliberately, because the failure mode we most want to avoid is people trusting it more than they should.
Cost and commitment
- How much does it cost?
- It depends on how many people use it and what you need it to do, so we scope it rather than publish a figure. What we will say plainly: it is a capital purchase plus a support arrangement, not a per-seat subscription. It does not increase when you hire, and it does not increase when your team uses it more. You get a written scope and a fixed price before committing to anything.
- Why is there no pricing on the site?
- Because a number without a scope is misleading. A system for four people and a system for thirty across two offices are not the same purchase, and quoting the smaller one to make the page look attractive would waste your time and ours.
- What if we decide against it?
- Then nothing happens and it has cost you a conversation. The assessment carries no obligation. We would also rather tell you honestly that you do not need one than sell you hardware that sits underused — a bad fit generates a refund request and no referrals, which is a poor trade for us as well as you.
- Are we locked in?
- No. The models are open and the software is standard. If you want to take support in-house, relocate the system, or have another provider maintain it, you can. We would rather earn a renewal than trap you into one.
Compliance and risk
- Does this make us Privacy Act compliant?
- No, and be wary of anyone who says otherwise. It removes one specific risk — the cross-border disclosure of confidential information to AI providers — which happens to be the one that is hardest to control by policy alone. Your security, consent, record-keeping and retention obligations are entirely unchanged.
- Do we still need an AI policy?
- Yes. The system changes what the policy has to say, not whether you need one. It is much easier to write a policy describing how to use the sanctioned tool than one prohibiting everything and hoping.
- What about security of the machine itself?
- A fair question and one people should ask more. It is a device on your network and needs to be secured accordingly — physical access, network segmentation, access control, updates. We cover this at installation and work with your IT provider. Moving data in-house is only an improvement if in-house is actually secure.
- Can we prove to a client that their data stayed onshore?
- You can demonstrate the architecture — the system is in your building, on your network, with no external connection required for it to work. That is a materially stronger position than pointing at a vendor policy, and it is something you can show a client in person.
Practicalities
- Where does it physically go?
- A server cupboard, a comms rack, or a shelf in a locked room. It is a compact unit, roughly the size of a small desktop, and runs quietly on a normal power point.
- Do we need IT staff?
- No. We configure, install and support it. If you already have an IT provider we work alongside them rather than around them.
- What happens if it fails?
- Support and warranty terms are scoped and confirmed in writing before you commit — including how quickly a failed unit gets attended to. A vague answer here is worth nothing, so we put it on paper.
- Do you only work in Perth?
- We are based in Perth and that is where we started, which means WA firms get the most direct service. We are talking to firms elsewhere in Australia — get in touch and we will tell you honestly what we can support well from here.
No obligation
Find out whether this fits your firm
Tell us how your team is using AI today and what you cannot afford to have leave the building. We will tell you honestly whether an on-premise system is the right answer — including when it is not.