What we do
Decide what to own. We build it and run it.
From a first workshop to AI you own, and the workflows and training around it.
The four steps
One step at a time.
Workshop
We sort how you use AI into what to keep renting and what's worth owning.
AI Ownership Workshop · No cost
Operations, IT and finance in one room for about 90 minutes, at your office or by video call. Bring two or three jobs that take too long. You leave with a one-page ranking: what to rent, what to own, a rough cost and a first pilot. Send Ty your answers from Find your setup first and you also get a short brief, with a page each for finance, IT, risk and operations, in plain language.
Map
We work out what your AI costs now, what it will cost once AI agents are running, and which jobs could move to a model you own.
Inference Audit
We price each job two ways, rented and owned, at today's use and once AI agents are running, and shortlist the models that suit it, with where each one comes from. We also read each AI vendor's terms: who keeps your data, whether you may reuse what their models write, and how you'd leave.
Prove
We run one job on a model you own and test it on your own work, before you commit to more.
Ownership Pilot
A model you own runs one job beside what you use now, on your own real examples. We record quality, time spent checking and cost per task before the pilot and during it. At the end you decide: keep it, change it or stop.
Build and run
We build the model you own, install what it runs on and keep it running.
Private Model Build and Managed Inference
- Private Model BuildWe choose a model, connect it to the documents it should look things up in, train it on your terms and formats where that helps, test it on your own examples and run it.
- Private Model ServiceA single entry point for every team and AI agent. It sends each request to the right model, rented or owned, records who asked what, and keeps spending within your limits.
- Hardware, supplied and supportedA small edge AI box for a site, a desk-side AI computer, a server in your comms room, or space in an Australian private cloud. We size it from your numbers, then install and set it up.
- Managed InferenceMonitoring, model updates, further training, support, and a monthly report on what you own and what you rent, on terms agreed in writing.
- AI workflows and agentsThe jobs AI does for you, such as tenders, documents, scheduling and site procedures. Each one runs on whichever model suits it, rented or owned.
- TrainingWe train your team on its own tasks and hand over the instructions, examples and review checks.
At the end you own your data, the records of what the AI did, the test sets used to check it, the trained model and its memory (the library of your documents it looks things up in), all listed in a register you keep.
AI workflows and agents
AI that takes on the routine work.
An AI agent is software that does a set job, such as gathering information or drafting a report, and passes anything important to a person. People keep the decisions.
- Drafts from your own work
- Tenders, reports and routine documents, drafted from your past work and templates for your team to check.
- Less rekeying and chasing
- Information moves between the systems you already use, and handoffs, approvals and follow-ups reach the right person.
- Answers you can check
- Your team finds the right policy, contract clause or record, with its source.
- Gaps seen before you sign
- Payment dates, obligations and capacity side by side, so a cash-flow gap or a clash shows up early.
AI agents
Two AI agents at work.
Each agent does a set job and passes the decisions to a person. Both demos use sample scenarios and made-up data.
A draft estimate, with its sources.
An AI estimating agent reads a sample tender, finds similar past jobs and current rates, and hands a draft to an estimator to check.
AI-narrated demonstration · A sample tender with made-up data.
Read the narration transcript
This is Gauge, a construction estimating agent built by Airon Systems.
A tender pack has just landed.
Gauge reads all 84 pages and indexes them against past project history.
It prices from current rate cards and checks how similar jobs actually delivered.
Where it sees margin risk, latent conditions, compressed programmes,
it flags it and asks for human judgement.
Then it drafts an evidence-backed estimate, every line traceable to a source.
An estimator signs off before anything goes out.
That's the design.
A site incident, with people making the calls.
An AI safety agent reads alerts from a sample site, asks a person to make the call, then runs the approved response and drafts the incident record.
AI-narrated demonstration · A sample scenario with made-up data.
Read the narration transcript
This is Marshal, a safety and incident response agent from Airon Systems, working behind video analytics on the site's existing cameras.
A PPE alert fires. In these conditions, Marshal reads it as early heat stress and asks for a human call.
The watch escalates. A first aider is already pre-positioned.
Person down. One human decision. And Marshal runs the response workflow. Dispatch, notifications, and incident record building in real time.
Standing rules carry the routine steps. People keep the judgment calls.
Afterwards, the incident record, corrective actions, and a full audit trail. Drafted from the event log, not from memory.
How we design AI agents like this
- Clear jobs for each AI agent
- Each agent is designed around a specific task. A coordinator passes work between agents and keeps the evidence together.
- Review and exceptions
- Gaps and uncertain answers are flagged, and important decisions go to the person responsible.
- Records you can check
- It only uses the data you’ve agreed to, and keeps sources and approvals together so anyone can check them later.
Training and support
Your team keeps using it.
- Training uses your own tasks and real examples, so people learn on the work they do.
- You keep the instructions, examples and review checks, so the team can carry on without us.
- After launch we keep it working as your people, processes and tools change.
AI Ownership Workshop
Find out what to rent and what to own.
About 90 minutes at no cost, and you leave with a one-page ranking, a rough cost and a first pilot.