Automate the work a machine should already be doing.

Software that does a repetitive office task end to end — reading an email, pulling the numbers out of it, putting them where they belong, and telling somebody when it cannot. It is not a chatbot bolted onto your website, and it is not a strategy document nobody acts on. It also does not mean replacing the systems you already run: most of what we build sits between them and moves work from one to the next.

An agent is a program given a narrow job, the access it needs to do that job, and rules about when to stop and ask a person. Book the delivery. Draft the reply for approval. Flag the invoice that does not match its order. It is not a general intelligence, it is not a member of staff, and it has no judgement outside the task it was built for. The narrower the job, the better it works.

There is no price list on this page, and that is deliberate rather than coy. Until we have sat with your team and counted where the hours go, any figure would be a guess dressed up as a quote — the same task costs very different amounts depending on how many systems it touches and how clean the data in them is. The audit is what makes an estimate worth anything, so it is scoped and priced on its own, first. Everything after it is quoted against what the audit found, before any of it is built.

The audit comes back as a written report. The pilot after it is deliberately one narrow task rather than a programme, because something small and running beats something large and half-built. It runs beside the manual process at first, so you can compare the two before anybody depends on it. We would rather give you a date on the call, measured against your own systems and the access we would need, than publish a timeline here that assumes both.

It depends on the model and on where it is deployed, and you are told which before anything is built rather than after. Some models run on infrastructure in South Africa, some are hosted abroad — commonly in the European Union or the United States — and some can run on hardware you own, which is the option for records that cannot leave the building. Where information does cross a border, section 72 of POPIA sets the conditions it may do so under, and our privacy policy sets out the ones we rely on.

Personal information you hold about your own customers stays yours: you are the responsible party for it, and anyone processing it on your instruction, us included, is your operator. Automation does not change that. What it does change is how many systems the information passes through, which is why the audit records every place a task touches personal information rather than only the places it saves time. Where a decision significantly affects somebody, section 71 of the Act expects a person to make it, so those stay with a person by design.

It will, and the question worth asking is where it gets caught. Every workflow is built with a point at which a person sees the output before it becomes a fact — the invoice posted, the message sent, the order placed. Steps that are low risk and easily undone run unattended; anything that spends money, contacts a customer or cannot be reversed waits for an approval. When it is wrong, we would rather it were wrong in a queue somebody reads than in your ledger.

You do. The prompts, the workflow definitions, the configuration and the documentation are yours, handed over in a form another firm could pick up and carry on with. Nothing is held in an account you cannot get into, and there is no licence that stops working the month you stop paying us. If you want to run it in-house or hand it to somebody else, we will hand it over — the same arrangement as the domains and hosting we run, which are in your name from the start.