You already see what AI can do
for your business.
Now take it to the next level — AI designed, structured and owned in-house, built around the way your business actually works.
Built so you can show anyone — your board, your accountant, your lawyer — exactly how it works.
The next step is
owning one.
A specialised agent, shaped to your niche. Open-source models, your data, your hardware or private cloud — and you own what comes out of it.
Generic AI, rented by the seat
In-house AI design and structure
Pick a trade. See what its agent takes off the desk.
Same idea, different shape. The agent learns one business deeply instead of all of them shallowly.
The practice buried in month-end
Reconciliations, queries and the same eleven client questions, every single cycle.
A ledger agent that knows your clients’ books
The lender whose file is the product
The core system records what was decided. Nothing holds why — the policy version that applied, the memo, the correspondence — so every question about a past decision starts with somebody reading a file again.
A file agent that draws on your own policies and decisions
The lead drowning in RFIs
Drawings revise, specs conflict, and the answer is buried in a 400-page document somebody emailed last March.
A project agent across drawings, specs and correspondence
The team answering the public, on a statutory clock
Permits, complaints, official information requests and papers for the meeting — all on a fixed headcount, all discoverable, all with a deadline somebody else set.
A public-service agent on sovereign infrastructure
The broker whose value is buried in the file
Renewal season, a claim in dispute, and the answer to both sits somewhere in eleven years of correspondence, schedules and endorsements nobody has time to re-read.
A file agent that draws on your own wordings and claims
The practitioner billing for reading
Discovery bundles, precedent hunts and first drafts eat the hours you would rather spend on strategy — and clients feel the bill.
A matter agent that draws on your firm’s own precedents
The quality manager who is really a librarian
The audit is in three weeks. The evidence exists — batch records, supplier certificates, corrective actions — spread across a shared drive, a filing cabinet and one person’s memory.
A quality and traceability agent built on your own records
The clinician typing after hours
Notes, referrals and recalls follow you home. The clinical work was the easy part.
A clinic agent inside your own four walls
The agent who lists, shows and chases
Six live listings, forty warm leads, three contracts mid-flight — and the follow-ups that never happen are the ones that cost the commission.
A listing and pipeline agent that never forgets a lead
The recruiter screening at midnight
Three hundred applications, nine roles, and every candidate expecting an answer.
A pipeline agent built on the roles you’ve actually placed
The operator quoting from the van
Quotes at 9pm, job sheets on the dash, invoices a fortnight late. The work is booked; the paperwork isn’t.
A dispatch agent that runs the paperwork
The operator who can’t invoice until the paperwork lands
The truck ran, the freight arrived, the customer is happy — and the invoice sits waiting on a signed docket somebody photographed in a yard three days ago.
A dispatch and billing agent built on your own runs
The network that has to prove it did the work
The assets are in the ground and the regulator wants evidence. What was actually done sits in field notes, contractor reports and photographs, in three formats and four places.
A records agent built on your own asset and field data
Four steps from conversation to a working colleague.
Map the week
A working session on where the hours actually go. We leave with a ranked list of workflows, an honest read on which are worth automating first, and an agreement up front on what the agent will never decide on its own.
Design the agent
Scope, data sources, tools and guardrails. What it can see and what it can act on — written down before anything is built.
Build in-house
Deployed on your own infrastructure, connected to your systems, built on your documents and evaluated against real cases from your business. Decision logging and human sign-off are built in, not bolted on.
Hand over and grow
You get the Handover Pack: a model card, a data provenance sheet, the decision-log schema, the never-list as built into the software, and a plain-English operating guide. Your team runs it from there — we watch its health and usage metrics, never your documents, tune it, and add the next workflow when you’re ready, or hand the keys over entirely.
All in-house. Nothing leaves unless you say so.
Where it physically runs is your call, driven by your IP and compliance needs. Whichever you pick, it connects securely to the software you already use, always behind a boundary you control.
Your system is yours alone. It is built on your data — your documents, your work, your examples — and nothing from your business ever improves another client’s system. At handover you get a data provenance sheet: what it was built on, from when, and what it wasn’t.
On your own hardware
Full sovereigntyA machine in your own office, sized to the firm. Best for regulated records and sensitive IP.
Data sits on your premises. Reachable only from your network, by people you already trust with the files.
Private rack, local data centre
In-jurisdictionDedicated hardware in a facility near you. Resilience and uptime without a hyperscaler’s shared tenancy.
Data stays in New Zealand, on hardware leased to you. Facility staff hold physical access; logical access is yours.
Dedicated cloud
Fastest to scaleIsolated cloud infrastructure when demand is spiky or teams are distributed. Same architecture, portable by design.
Data sits in a single-tenant instance in a region you pick. No shared model, no shared storage, keys held by you.
The shape of it, and exactly where the line sits.
Open models, your documents kept live, a person in front of anything that leaves
What you own
What we retain
If we part ways, the system keeps running and someone else can pick it up. That’s the point of building it this way.
Where it runs, and what it costs to keep running.
Two questions every owner asks once they’re past “does this work”. Here are the honest answers.
Sized by the size of your firm, not by a spec sheet
What a cloud-only subscription commits you to
Start where it makes sense for your business.
Every package is scoped to the business type. Most start with the audit and grow from there.
AI Audit
Build & Install
Managed & Grow
The monotony leaves the business. What you do with the hours it hands back — more work, more play, more time at home — is entirely your call.
Isn’t building our own AI a big, expensive project?
Smaller than most expect. Open-weight models are free to license and now run well on a single workstation-class machine. The cost is in the design and integration work — scoped as a fixed-price build, typically weeks rather than quarters, starting with one workflow that clearly hurts.
Do we need an internal AI team to run it?
No. We deliver it documented and monitored, and either support it on retainer or hand it over with training.
What if we outgrow it, or the models improve?
That’s the point of open weights. Models are swappable components; the value sits in the structure, data pipelines and evaluations around them. Upgrading the engine doesn’t mean rebuilding the car.
Is using AI going to get me in trouble with the Commerce Commission?
The short answer is that a business stays responsible for what its tools produce, and the rules read the same way whether a person or a system made the call. The exposure sits with tools that act on their own, or that pass data and signals between competitors. Everything we build recommends rather than decides, and keeps a log of who approved what — so if you are ever asked, you can show the answer rather than argue it. Further reading: the Commerce Commission. We build systems that are straightforward for your own lawyer to review; we don’t give legal advice.
What if my competitor uses the same AI vendor as me?
That is the real risk with shared, multi-tenant tools — one algorithm learning from several competitors at once, inside a vendor you both pay. We build one model per client, on your data only. Nothing from your business ever informs another client’s system.
Do I need my own hardware?
No — but you can, and plenty of firms should. What it takes, sized by the size of the firm, is set out in what it takes to run above, along with the case for owning rather than renting.
My industry isn’t on your list. Does that matter?
Not much. The list shows shapes we have already thought about, not the limit of what this works on. What decides it is whether your business runs on documents you own — files, correspondence, records, a history of how you do things — and whether the same questions get asked of them every week. If that is true, the method is the same and only the vocabulary changes. If your work is mostly serving customers over a counter, or it already lives in one tidy system that answers well enough, we will say so on the call rather than after the invoice.
Where should we start?
The free 15-minute call. We map your week, pick the one workflow with the best ratio of pain to complexity, and tell you honestly whether it’s worth building yet.
Fifteen minutes. One honest answer.
Tell us what your week looks like and we’ll tell you whether a specialised in-house agent is worth building — and roughly what it would cost. No deck, no obligation.
The booking page is Cal.com’s. Your name and email go to them to make the appointment, and reach us on the invite. This page itself collects nothing — no form, no cookies, no analytics, and nothing loaded from anyone else.