Knowledge & ops agents
Back-office automation, built to your business.
The invoices, the data entry, the retyping between systems, handled by an agent built to how your business already works. Only the exceptions reach you.
Built on your rules · the odd one comes to you
What it takes on
The work nobody wants to do
Not the decisions. The repetitive paperwork around them, handled the way your business already works.
Invoices, read and posted
An invoice arrives as a PDF or a photo. The agent reads the vendor, the amount and the due date, checks them against the order, and posts it. The ones that do not match wait for you.
Stock and orders, in sync
An order in one system, a stock count in another, a supplier due a reorder. The agent moves the numbers across, updates the counts, and flags the reorder, so the supply chain stops living in a retyped spreadsheet.
Answers from your own files
The answer is buried in a contract or last quarter's report. The agent reads them and answers in a line, in your words, with the source named beside it.
How it works, and stays safe
A script clicks. An agent decides.
It reads what actually arrived, checks it against your rules, and acts across your tools. When it is not sure, it stops and asks.
PDFs, scans, a photo of a receipt, an email thread. It works from what arrived, not a template one odd layout breaks.
Against the rules in the spec you approve, shaped to how your business works, not a generic tool you bend yourself around.
Below a confidence line it posts nothing. The exception waits for a person, and nothing binding goes through without one.
Not listed?
Looking for something else?
These are the shapes back-office work usually takes, not a menu. If your team's tedious work is something else, tell us. Twenty minutes, no pitch.
FAQ
About back-office agents
What people ask before they hand the paperwork over.
Optical Character Recognition (OCR) turns an image into text and stops. Robotic Process Automation (RPA) clicks through a fixed sequence and breaks when the screen changes. An agent reads the document, decides what it means against your rules, and writes the result into the systems you already run, and it asks a person when it is unsure. It acts, rather than only reading or only clicking.
Accuracy is set by a confidence line you agree, not left to chance. Above it the agent acts; below it it stops and puts the case in an exception queue for a person, with everything it read attached. It is built to fail in one direction: when in doubt it asks, it never guesses. Typically only a small share of cases need a human, and that share is visible from day one.
Almost always, and usually as a configuration question rather than a rebuild. The agent reaches your tools over MCP and n8n's connectors, which cover several hundred services, and we build against the accounting or ERP you already pay for, whether that is Bexio, Abacus, Xero, QuickBooks or SAP. We name the exact connections in your specification once we have seen the account, so the answer is a commitment and not a maybe.
Yes, because it answers only from your documents, not from the open internet, and it names the source beside every answer so you can check it. When your files do not hold the answer, it says so instead of inventing one. That grounding, and the citation next to it, is what keeps it honest.
On a server you choose, in your own accounts, and only your people have access. The agent runs inside your environment rather than on ours, so your documents and records never sit on a system you do not control. We hold no copy, and our access comes off at handover.
Compliance is built into the specification, not promised afterwards. For an EU or Swiss business the agent is built to the GDPR and the FADP: personal data is minimised, scrubbed where it is not needed, and kept in a region you choose. If you operate under another country's law, we build to that law instead. The rules that apply to you are written into the spec before anyone opens an editor.
Weeks for a first agent, and no. It runs on your own accounts and credentials and does not check in with us to keep working. There is no standing fee holding it alive. Training is included so your team can operate it, and if you want a change later you ask, and we price that change on its own.
The judgment. The agent handles the repetitive middle, the reading and checking and posting, and a person handles the exceptions, the approvals, and anything binding. You decide where that line sits before it ships, and you can move it as trust builds. The point is to take the dull work off people, not the decisions.