Perimeter · the sovereign AI control plane
Your team’s AI, running work you can watch happen.
Perimeter runs your assistants, your document search and your team’s AI-driven work on infrastructure based in Newcastle — with a verifiable record of every action it takes, and a way to undo any of them.
What’s verifiable, not claimed
- Every action is logged
- A hash-chained, checkpoint-signed audit ledger. Export it yourself and verify it offline with one command — no trust required.
- Agents can’t run wild
- Every tool call runs inside a WebAssembly sandbox with an explicit, named permission: which host it may reach, which folder it may write. Nothing implicit.
- Mistakes are reversible
- One-click rollback of a run’s file effects, backed by a full, deterministic replay log.
- Independently checked
- 52 of 52 platform requirements verified by independent verifier agents before this page went live.
How it works
Three steps. Every one of them on the record.
Your documents come in.
Upload contracts, records, notes. They’re indexed on Newcastle-resident hardware and never sent to a third-party API.
Your assistant answers, with receipts.
Every answer cites its source document. Every action — search, draft, write — runs sandboxed with a named permission and lands in the ledger.
You can check, export and undo.
Export the ledger and verify it offline, any time. If an agent run did something you didn’t want, roll it back.
Who it’s for
Firms with a confidentiality obligation and real work to do.
- Law & conveyancing
- Accounting & bookkeeping
- Defence supply chain (DISP members)
- Engineering & IP-heavy firms
- Health & allied health
- Anyone bound by a confidentiality obligation
Pricing
Start small. Keep the report either way.
Prices include GST. Quoted in writing before you commit to anything.
Sovereign AI Diagnostic
$2,200
A paid, fixed-fee, 2–3 day technical review — not a form, not an AI-generated audit. We sit with your team, map where sensitive data actually goes today, and hand you a written report: a data-sovereignty risk map and a fixed-price quote to fix it. Keep the report even if you go no further.
Book the diagnosticFour-week guaranteed pilot
$12,100
One assistant, your documents, up to 15 people, running on Newcastle hardware — working in four weeks or the diagnostic fee is refunded. We take the delivery risk, not you.
Ask about the pilotFair billing: cost + 20%
first $5/mo free
No tiers. You pay exactly what your usage costs us — per model, per token, shown to the cent on your usage page — plus one openly disclosed 20% margin. Every account starts with a free monthly allowance. Consultancy (how to use it, running it at enterprise level) is quoted separately, so the usage bill stays honest.
Ask about fair billingThe four-week guarantee.
If we don’t have a working pilot running on your workflow within four weeks of kickoff, the diagnostic fee is refunded in full — no argument, no fine print.
Developers
A drop-in API for open-weight models, hosted in Newcastle.
Same request format your existing SDKs already speak — but the models are open-weight (Qwen, Llama-class) running on our hardware in Newcastle. No OpenAI, no overseas API, nothing leaves Australia. Per-tenant keys, rate limits and monthly budgets; usage and an exportable audit ledger built in. Keys are issued by hand while we’re in early access — tell us what you’re building.
curl https://perimeter.newcastlerising.com.au/v1/chat/completions \
-H "Authorization: Bearer pk_…" \
-H "content-type: application/json" \
-d '{"model":"qwen3-8b","messages":[{"role":"user","content":"Hello"}]}'Questions
Asked often, answered straight.
Where does our data actually live?
On infrastructure based in Newcastle, NSW. Nothing sensitive is sent to an overseas API.
Does this meet DISP / Essential Eight requirements?
Perimeter is built to support DISP member obligations. Ask us for the current control mapping for your membership tier — we’ll tell you plainly what’s covered and what your organisation still owns. We don’t claim certifications we haven’t earned.
What happens to our documents if we leave?
You can export everything — documents, ledger, usage history — at any time, in open formats. No lock-in clause, no delay.
Can we leave the subscription at any time?
Yes. Month to month after the pilot; no minimum term, no cancellation fee.
Does anything train on our data?
No. Never. Models are used for inference only.
Who can see our documents and conversations?
Your team, isolated from every other customer. We have access only for support, and every access we make is itself logged in the ledger you can export.
What’s the “glass-box agent” thing, practically?
When your assistant does more than answer — searches a folder, drafts a file, writes something — it runs inside a sandbox that can only do what it’s explicitly allowed to do, and every action is recorded so you can see exactly what happened and undo it if needed.
What if the pilot doesn’t work for us?
Covered by the four-week guarantee: the diagnostic fee comes back, no argument.
Do we need our own hardware?
No — we host it on Newcastle-based infrastructure. If you’d rather run it on your own premises, that’s a build option too; ask at the diagnostic stage.
How is this priced?
Fair billing: you pay what your usage costs us, plus one disclosed 20% margin, shown to the cent on your usage page — no tiers, no seat counts, and the first $5 a month is free. Consultancy and the pilot are fixed-price and quoted in writing. Nothing is charged before you agree.
Newcastle NSW · Built and operated here
Talk to the engineer who built it.
A thirty-minute conversation is enough to tell whether the diagnostic is worth your time.