Start with the truth.
Any recurring job that eats time can be automated — quoting, invoices, stock, inbox, bookings, reporting. Most people start with the audit; some already know exactly what they want built.
Automation Audit
Find the time you're losing. Already bought AI or automation that doesn't deliver? We audit it, fix it, or replace it — and show you what your manual tasks actually cost.
How it works
A fixed-fee audit. We sit with your team, map what they actually spend time on, and turn it into numbers you can act on.
- Task inventory — every recurring manual task, who does it, and how long it takes. No guessing; we measure.
- Cost-of-doing-nothing math — what each task costs per year at loaded wage rates. The number is usually uncomfortable.
- Prioritised automation roadmap — for each task: automate it with a pipeline, hand it to an agent, or leave it alone. Each item comes with its own payback period.
- Existing tooling review — already paying for AI or automation that underdelivers? We tell you honestly whether to fix it or kill it.
Why an audit first. Because most "AI projects" fail the other way round: someone buys a tool, then hunts for a problem. We start from the tasks your staff shouldn't be doing and work forward. If automation doesn't pay for itself in months, we say so — and you keep the audit document either way.
Fixed fee, quoted before we start. If you go ahead with a build from the roadmap, part of the audit fee comes off the project.
What we automate
The jobs that come up most. Every one is a version of the same thing: work a person is doing because nobody connected the systems yet.
Quoting & follow-up
AutomationEnquiries land in one place, get qualified against your rules, and the quote goes out with a follow-up schedule attached. Nothing sits in an inbox for three days, and nothing goes cold because someone was on site.
You stop: copying details into a quote template, and remembering who to chase.
Invoices, purchase orders & dockets
AutomationSupplier invoices, delivery dockets and purchase orders are read, matched to the right job or supplier, and entered into your accounts. The paper still arrives; nobody types it twice.
You stop: retyping line items into Xero, Sage or QuickBooks.
Stock & listings
Live proofPhotograph the item and the system removes the background, reads the title, platform, condition and barcode, checks the market price, and queues it for approval. Approve it and it is live on the storefront.
You stop: typing titles, cropping images and pricing by feel. See it running →
Shared inbox & replies
AI-assistedMail is classified as it arrives — quote request, complaint, supplier, spam. A reply is drafted in your tone with the right details pulled in, and left for a person to read and send.
You stop: reading every email twice, and retyping the same answers.
Bookings & reminders
AutomationSelf-serve booking that respects your real availability, with confirmations, reminders, deposits where they make sense, and a record of no-shows.
You stop: phone tag, double-entering the diary, and chasing no-shows.
Reporting & alerts
AutomationYour numbers assembled overnight and delivered as one page in the morning: sales, stock, jobs overdue, anything outside its threshold. Exceptions get alerted; the rest waits until you ask.
You stop: rebuilding the same spreadsheet every Monday.
Not on the list? That is usually the interesting one. If a person does it the same way every week, it can almost certainly be automated — the audit is there to tell you whether it is worth it.
How we build it
- Smallest useful version first. The part that removes the typing ships first; the clever parts come after it has earned its place.
- Human approval, every time. Every automation starts with a person approving its actions. Autonomy grows only as trust does, and you can always take the trigger back.
- It runs on your systems. Your accounts package, your inbox, your site, your database — not a walled garden you have to move into.
- Monitored from day one. The same Prometheus and Grafana stack we have run for data centres and customer infrastructure for years. When something misbehaves, we know before you do.
Local AI & custom agents
When the data cannot leave the building, or the job needs judgement rather than rules, we go deeper than scripting.
- On-premise LLM stacks — your own AI on your own hardware. We run a multi-GPU local inference server in-house with 262K-token context, custom quantization and scheduling, and we have built our own Android chat app that runs local models with voice and streaming.
- Agents & MCP servers — AI that can use your tools: email, databases, your own software. Built custom, because your business is not a template. We have built our own MCP server and agent platform from scratch, including MCP servers on edge hardware that control real devices over natural language.
- The marketing agent, live — a new listing triggers a blurb, the owner approves it in Telegram, and it posts to Instagram and Facebook. It runs at aussieretrogames.com today.
How it's priced
- Audit — fixed fee, quoted before we start. Part of it comes off the build if you go ahead.
- Builds — quoted per project after a scoping call. Scope in writing, no surprises. Local AI spec, build and tuning work is available at a day rate.
- Care plan — from €500/month once it is live: hosting, monitoring, changes and support. You are not left holding something nobody can maintain.
What we don't do
- Staff augmentation. We are not a body shop; we build the thing that removes the work.
- 24/7 phone support. Monitoring is continuous, but we are a small studio and we say so up front.
- Reselling licences. We are not here to put you on a subscription for software you could buy yourself.
Not sure which you need?
The audit answers that — that's its job. Bring the task that annoys you most and we will tell you whether it is worth automating at all.