AI agents that know when to stop.
Agents that read, decide and act across your systems, with a person approving what matters. GDPR- and EU AI Act-ready from the first commit.
Where agents earn their keep.
Repetitive work with clear rules and real volume. These are typical examples, not a menu.
Document intake
Read invoices, claims, contracts and forms, extract the fields and route them to the right system.
Support triage
Classify incoming requests, draft replies from your knowledge base and hand the hard ones to a person.
Back-office reconciliation
Match payments, orders and records across systems and flag what does not add up.
Knowledge assistant
Answer staff questions from your own documents, with sources, inside your security boundary.
Compliance monitoring
Check documents and processes against your rules and prepare evidence for audits.
CRM and data hygiene
Enrich, deduplicate and update records so your team stops doing it by hand.
How we build them.
Start with one workflow.
One measurable process, end to end, before anything else. It proves value fast and shows where the risks are.
People approve what matters.
You set the thresholds. Anything above them waits for a person, with the evidence laid out.
Tested before launch.
An evaluation set built from your real cases, including the awkward ones, run before every release.
Runs where your data lives.
In your EU cloud region and accounts, with logging and access controls you can audit.
Our experience includes production conversational AI for a US insurtech.
Anatomy of an agent.
Six steps, every time. The model does the reading and the routine work; your rules and your people decide what happens next.
Trigger
An email, a form or a file arrives.
Understand
The model reads it and pulls out what matters.
Act
It works in your systems: ERP, CRM, ticketing, documents.
Check
Rules and thresholds you set decide what is safe to do alone.
Approve
A person signs off on anything above the line.
Log
Every step is recorded, so you can audit any decision.
Transparency duties for chatbots already apply under Article 50 of the EU AI Act. Check your system.
By industry
Describe one workflow you want automated.
What comes in, what happens to it today, and what a good outcome looks like. We reply within 24 hours on working days.
Which models do you use?
Whichever fits the task and your data rules: hosted models such as Claude or GPT through EU endpoints where available, or open-weight models running in your own cloud.
Will our data be used to train models?
No. We use provider settings and contracts that exclude training on your data, or models you host yourself.
How long until something is running?
Discovery takes a week. A first working version of one workflow typically follows within a few weeks, scoped in the discovery plan.
Is an AI agent a high-risk system under the AI Act?
It depends on what it decides. Most back-office agents are not high-risk, but agents used for hiring, credit or insurance pricing are. The free checker gives you a first answer.