Files that handle themselves
Most companies have one person who knows where the files belong. That is not a system —
that is a colleague who remembers. We build the system: files are read, named, sorted and
archived according to rules that fit your work.
Where the files come from
- Attachments in emails — invoices, order confirmations, delivery notes.
- Scans — from the copier or scanner, often with a meaningless file name.
- Exports from business systems — CSV or PDF extracts that need to go somewhere else.
- Files that simply arrive — downloads, shared folders, an external party uploading.
File received→
Opened and read→
Named as agreed→
Filed in the right folder→
Recorded in the log
How the system knows what the file is
The content is read — not just the file name. From that and from your rules (customer,
date, document type, department) the file is placed. If the system is in doubt, it puts
the file in a for review folder and says so, instead of guessing. And nothing is
deleted — it can always be found again.
What you decide: which folder structures are the right ones and how
the files should be named. That is your professional judgement, not ours. We write it
down as rules, so that two people no longer do it differently.
We operate the programs — from the outside
Here is the most important difference from a traditional integration. We do not connect
to your program's database, we install no connector, and we change nothing inside the
program. Instead we operate the program on screen: we open it, click, type in the fields
and collect the result — exactly as you would do it yourself.
Simple, surface-level and visual handling — exactly as you would do it
yourself. That means it works with the programs you already have, including the
old ones. Also when the vendor offers no integration, and when there is no developer
available to build one.
What "from the outside" means in concrete terms
- No API access. We do not use the program's data interface — it is often locked, expensive or non-existent.
- No database access. We do not read your tables and we do not write to them.
- No changes to the setup. Nothing is installed into the program, and its version is untouched.
- No special permissions. The agent works with an ordinary user — it can only do what that user may do.
What we do not get access to: your data in the background, other
users' cases, your servers, your keys. What the agent can do is what an employee with
the same user and the same screen can do. No more.
What it requires from you
- The program is installed (or reachable in the browser) as it is today.
- There is a user the agent may work with — and agreement on what it may do.
- A machine the solution can run on. We are happy to run on a machine at your premises.
Where the limit is
Because we work on the surface, we depend on the screen looking as expected. If the vendor
changes the layout, the agent notices and stops — it does not carry on guessing. We are
notified, update the recipe, and run again. And the system is not faster than the program
itself allows: we operate it as a human would, only without getting distracted and without
having to remember the order.
Documents prepared automatically
Many tasks are not "process a file" — they are "open the program and finish a document".
An invoice that has to be entered. An order confirmation that must be produced as a PDF and
sent for approval. A template that needs the same details as last time.
Task received→
Details read→
Program opened→
Fields filled in→
Saved in the right format→
Ready for approval
The work we typically take over
- Entry in the accounting or invoicing system from a receipt or an email.
- Filling in templates — Word, Excel or a business system — with details from another source.
- Converting and naming — the same document in the format and with the name the recipient expects.
- Assembling drafts — the document is ready when the employee arrives in the morning.
The draft is never sent or booked by itself. It is placed ready, with a receipt of what was
done, so a human can approve it at a glance.
Agents that read and prepare
An agent is not a magical being — it is a piece of software that can read a task, make a
small decision and carry out several steps in a row. That is exactly what makes it useful,
and exactly why it belongs on a short leash.
We build agents that read, sort and prepare: read an enquiry and find the
details that are needed, sort a pile of files by content, find the receipts a case is
missing, and lay the work ready for a human. The decision stays with you.
- No agent gets free rein from day one. Autonomy is earned through documented accuracy and withdrawn at the first bad step.
- It stops when in doubt. If it does not recognise the situation, it asks — it does not guess.
- Everything can be verified. Every run, approval and deviation is recorded.
Read about the security and governance model →