Your critical data stays in your company. The model works 24/7.
Cloud AI is fine for a public brochure. It is the wrong default when the file cannot leave the building. We put a computer in your office, load a free local model, and it answers around the clock — no tenant, no transcript in America.
From a desk PC to an air-gapped node.
What we actually install
- Hardware you own. Sourced through Australian channels. NVIDIA, Apple Silicon or AMD — whichever fits the room, the power budget and the model size. Not a rental cloud GPU with a flag on it.
- Hardened OS. Disk encryption, local accounts, no consumer app store junk, outbound blocked except what you approve.
- Local model runtime. Typically Ollama / llama.cpp / vLLM plus Open WebUI. Open-weight models (Mistral, Qwen, Gemma, Llama and peers) chosen for the job, not the press release.
- Private RAG. Your procedures, briefs, contracts or drawing notes — indexed on the box. Nothing uploaded to train someone else’s model.
- Handover. Admin logins, a one-page runbook, and a workshop so the team uses it on real files the same week.
When on-prem is the right call
- Legal privilege or health records
- APRA-regulated or CPS 230 conversations
- PSPF / ISM / Essential Eight environments
- Defence industry DISC members and primes
- Boards that have already banned public ChatGPT
- Sites with poor or contested connectivity
When it is not
If the work is public web copy and the risk is low, a governed cloud tenant is cheaper. We will say so. Many firms run both: local box for the sensitive corpus, cloud for the rest.
How an install runs
If we put the box in, we stay on it.
An unpatched on-prem LLM is worse than no LLM. Desk Care A$490, department A$990, air-gap A$1,690 a month — health, patches, model drops, RAG reindex, restore test. Full Care list.
Straight answers
Does any prompt leave the building?
Not on the default install. The runtime listens on your LAN. Outbound is blocked. If you later want a hybrid (local for sensitive, cloud for public), that is a separate, written rule.
Is this as good as ChatGPT or Claude?
Frontier closed models still win on the hardest reasoning. On-prem wins on control, cost-at-volume and jurisdiction. Most teams use the local box for draft, search and routine work, and keep a sanctioned cloud path for the 5% that needs a frontier model.
Can you put Claude or GPT on the PC?
Those vendors do not hand over weights for a PC in your office. On-prem means open-weight models you are allowed to host. Closed models stay in an accredited cloud region if you need them.
Do you IRAP-assess the box?
We document controls (encryption, egress, accounts, logging) so your existing assessor can do their job. We do not sell a fake accreditation sticker.
Defence SECRET / TOP SECRET?
No. We work Defence-adjacent and departmental admin at OFFICIAL and help you walk Protected-adjacent tooling into your own accreditation. Classified capability belongs with cleared primes and ASD-assessed platforms.
Power and noise?
A desk LLM PC is in the same class as a quiet workstation. GPU nodes need airflow and a grown-up PSU. We size that before anything is ordered so it does not end up under someone’s desk next to a heater.
Tell us what cannot leave the room.
We will tell you whether a desk PC, a departmental node, or a governed cloud tenant is the honest answer.
Free local models. Your critical data never leaves your company.
Safe, on the box you own, working 24/7. Prompts, files and answers do not go to ChatGPT, Gemini or any other cloud. Gemma, Llama, Qwen, Mistral and Granite run here — night shift, public holiday, air-gap. No tenant. No meter running in California.
Ollama libraryLlama, Qwen, Mistral, GraniteMapped to reception, RAG, coding and government admin — not a raw model dump.
From A$8,400 + A$490/moWe install the boxHardware you own. Care pins versions when the library moves.
A technician sets it up in your city. You do not travel.
AI Evolution is national. Head office is Unit 16/87–91 Heatherdale Road, Ringwood VIC — the kit is specified here. Delivery and install is done by technicians from The Original PC Doctor: every capital city, plus regional benches including Newcastle, Wollongong, the Gold Coast, Geelong, Canberra and Darwin. Same people who already walk into Australian offices with a screwdriver and a police check.
RegionalNewcastle, Wollongong, Gold Coast, Geelong and moreIf PC Doctor already services the suburb, the LLM PC can be stood up there.
1800 953 560Book the city, not a flight to MelbourneWe confirm the local tech on the scoping call. Care stays with AI Evolution.
Highly scalable. Need more agents? No worries.
Train it once. When the work doubles you add another agent or another seat — not another labour-hire line. It grows as your business does. That is the unfair advantage. Take it before your competitors do.
Effortless growthIt grows as your business doesQuotes, follow-ups and intake keep pace. You hire for expansion, not to plug holes.
Unfair advantageGet it while the other firms are still writing a strategy deckFirst process live in weeks. Care keeps it earning.