← Edge Attention

How our contracts are structured

A custom model trained for your task, evaluated on your own data, and deployed to run inside your business. One fixed price between A$8,000 and A$20,000, about thirty days, and it is yours at the end. The technical scope that comes first is free.

This is built for organisations of ten to fifty people who hold material they cannot put into a public AI service. You do not need an IT department, a procurement process, or a budget cycle to work with us.

What we build

A model delivery can be almost anything that learns from your data. Typical examples:

Athlete tracking

Vision

Track players, poses, and movement from broadcast or sideline video.

Traffic detection

Vision

Count, classify, and monitor vehicles from roadside cameras or drones.

● try the live demo →

Energy anomaly detection

Telemetry

Catch abnormal patterns in grid, plant, or building telemetry before they become failures.

● try the live demo →

Deep research agents

Agents

Models that plan, search your knowledge sources, and write cited briefs.

● try the live demo →

General AI models

Language

Private assistants and chat models trained on your domain, running in your infrastructure.

Math and reasoning models

Reasoning

Models tuned for calculation-heavy and structured reasoning workloads.

Automated computer use

Agents

Agents that operate software the way your staff do: screens, clicks, and forms.

PII redaction and data safety

Safety

Find and strip personal information before documents leave your perimeter, for example ahead of offshore model calls.

● try the live demo →

AI guardrails

Safety

Screens that catch prompt-injection attacks and abusive content before text reaches your models.

● try the live demo →

Synthetic media detection

Safety

Screen incoming imagery, documents, and voice audio for AI generation, a first line against fabricated evidence and cloned-voice fraud.

● try the live demo →

Document intelligence

Language

Extraction, classification, and search across your paperwork and records, from Q&A to pulling tables out of statements.

● try the live demo →

Reading scanned archives

Language

Turn filing cabinets and scanned PDFs into text your people can search, without the files leaving the office.

● try the live demo →

What it costs

A$8,000–20,000

One custom model, trained, evaluated, deployed and handed over

Where you land in the range depends on how hard the task is and how much work the data needs, not on how long it takes us. You are quoted a single number after scoping and that is the number you pay.

The technical scope before it is free

We do not quote until we have sat with your team and looked at the data you actually hold. That first assessment costs nothing and commits you to nothing. If your data will not support what you are asking for, we will tell you that instead of selling you a model that cannot work, and you will have lost nothing but an afternoon.

The quoted price then covers the work. Two things come from your side, and we say so before the engagement starts rather than halfway through it.

What you provide

There are no surprises here. Both of these are agreed in writing during scoping, before any money changes hands.

Your data

The material the model learns from, collected by you. We will tell you during scoping how much is needed and what shape it has to be in. If the data is not there yet, that is the first conversation to have, and we will say so before you commit to anything.

The compute

Training runs on hardware you pay for, usually a rented cloud instance for the training window. We specify exactly what is required for your task and why, so you are not guessing or over-provisioning. For most deliveries this is a modest cost measured in hundreds of dollars, not thousands.

The thirty days

Every engagement runs to the same shape, usually twenty to thirty working days from first conversation to your staff using it.

  1. Days 1–3

    Technical scope — free

    We sit with the people who will actually use it, watch the work happen, and look at the data you already hold. You leave this step with a written definition of what the model must do, what "working" means, and an honest answer on whether your data supports it. This costs nothing and carries no obligation. Only after it do we quote.

  2. Days 4–15

    Build and train

    The model is built and trained against your material and measured against the success criteria we agreed. We tell you exactly what compute to stand up for the training window and why.

  3. Days 16–25

    Deploy and test

    It goes into your environment, on your hardware, and gets tested on real work by your staff rather than on a sample set.

  4. Days 26–30

    Handover

    Your people are trained on it, the documentation is written, and you own what is left behind. No dependency on us to keep it running.

What you are actually buying

The price is the whole price, and the model is yours when it is over.

One engineer, start to finish

The person who scopes it trains it, the person who trains it installs it, and the person who installs it teaches your staff. No account managers, no juniors learning on your money, no handovers where context gets lost.

Fixed price, agreed up front

The number in the contract is the number you pay for the work. No hourly billing, no variations for work that turned out harder than expected. If it runs long, that is our problem.

Evaluation on your own data

You get measured accuracy against a held-out set of your material, not a vendor claim or a benchmark from a paper. The success criteria are agreed in writing during scoping, and the numbers at handover are against those criteria.

It runs without us

The model runs on your machines and keeps working if we disappear. You get the weights, the deployment, and the documentation. No per-seat licence, no monthly fee, no calls home.

When it is bigger than this

Some tasks need a model too large to train on a single machine. Once training has to be distributed across multiple GPUs, it stops being a thirty-day engagement and becomes a different kind of project: longer, materially more expensive to run, and needing a delivery team rather than one engineer.

We will tell you at scoping if your task falls into that category rather than quoting you a number we cannot deliver on. Those engagements are structured separately, start in the low hundreds of thousands, and are worth having a different conversation about. Most work we see does not need it.

Want to see the quality of work before talking numbers? Our models run live in your browser.

Request a modelAvailable models