Files that handle themselves
When emails, scans or exports arrive, they are read, named, sorted and put in the right place. Nobody empties the inbox on a Friday. No file is called document(3).pdf.
Automation and AI · built in Hobro
We build small, everyday systems for Danish companies: files that sort themselves, documents that get prepared, and programs that get operated — as if a colleague were at the mouse and keyboard. Simple, visual, and without touching the inside of the programs you already use.
None of them requires access to the inside of the programs.
What we can build
Four kinds of systems we build. Each of them solves a task that somebody is doing by hand today.
When emails, scans or exports arrive, they are read, named, sorted and put in the right place. Nobody empties the inbox on a Friday. No file is called document(3).pdf.
We control your programs from the outside: we open, click, type and collect — exactly as an employee would. With no deep access to the program itself.
An invoice, an order, a template: the system opens the program, fills in the fields, saves in the right format and leaves the draft ready for approval.
Agents that read a task, sort the content and prepare the next step. Nothing is sent, approved or deleted without a human.
How it works
We always start with one task. If it works, we expand. If it doesn't, we take it out again.
One conversation where we follow the task that hurts. We write down what is done today, and how often.
You get a solution that does one task properly — not a big system. It runs alongside your normal work while it is being tested.
The system takes over once it demonstrably gets it right. We look at the runs, and you can see what it has done and what is waiting for approval.
Security and responsible AI
Because we operate the programs on screen rather than through their internals, we do not need access to your data, your servers or your permissions. What we cannot reach, we cannot break.
No agent gets free rein from day one. Autonomy is earned through documented accuracy — and withdrawn at the first bad step.
If the agent does not recognise the screen or the situation, it stops and asks. An unanswered question is not a permission.
Runs, approvals and deviations are recorded, so a third party can verify the work afterwards.
The models run locally on a machine at your premises. No customer data is sent to a cloud service — and the task is solved even when the connection is down.
Read more
Most AI solutions are an app in front of a remote service. We chose the other way: the agent runs on a machine at your premises, and nothing leaves the network.
About
HumAIn Labs is a Danish software company. The name is a play on words: human meets AI. We believe technology should make people stronger — not redundant. We only deliver something we have first tried out in reality ourselves.
Do you have a routine that eats time — or an idea that needs testing?