We develop AI agents for companies with their own processes, several systems to connect and strict rules on data. The agent answers from your documents using retrieval-augmented generation (RAG), acts in your ERP and CRM with the right permissions and leaves important decisions to a person. It runs in an EU cloud or on your own servers.
↓ Scroll
It looks good in a demo, but cannot read from your ERP or write to your CRM.
Contracts, customer data and projects cannot go to a service whose location you do not control.
An assistant that does not cite the document behind its answer is not usable in a structured company.
Without logs and tests no one knows how often it is wrong, so no one trusts it.
We will be the first to tell you if an off-the-shelf product fits.
the case is standard, data is not confidential, no integration with internal systems is needed and volumes are low.
it must read and write in ERP, CRM or document systems, data must stay in the EU or in-house, and you need per-user permissions, logs and testing before production.
Chosen for the case: provider APIs with EU data processing, or an open-weight model installed on private servers.
The agent searches your documents and answers from them, citing the source. Fine-tuning, i.e. retraining the model, is rarely needed: we propose it only when the case calls for it.
The actions the agent can take: read an order, create a draft, open a ticket. Each with its own permissions.
Human approval on important actions, a log of every operation, thresholds beyond which the agent stops.
Read and write via API or database, with dedicated service users.
Contacts, deals, activities and customer history.
Documents indexed with their original permissions.
Email, calendars, shared files.
Any system that exposes data in a controlled way.
An agent goes live when the numbers say so, not when the demo looks convincing.
Real cases with known correct answers, hard ones included.
Correct answers, hand-overs to a person, timings, cost per operation.
Attempts to push the agent outside its rules: off-topic requests, instructions hidden in documents.
Each release is compared with the previous one on the same test set.
We help on the technical side. For legal aspects, involve your own counsel.
An organisation using an AI system in its business is a deployer. Since 2 February 2025 it must take AI literacy measures for staff (Article 4).
Since 2 August 2026, if the agent talks to people, it must disclose that it is AI, unless obvious.
Obligations for high-risk systems in Annex III (employment, credit, education, biometrics) apply from 2 December 2027, after the postponement in Regulation (EU) 2026/1744.
Operation log, a named responsible person, a procedure to suspend the agent.
Provider APIs with EU data processing. Fast start, pay-per-use.
Model and agent on a dedicated server managed by us.
Everything on your servers. The assistant on our own website runs on Google’s open-weight Gemma model, on one of our servers.
Point us to a repetitive task: incoming email, documents, data entry. We check whether an AI agent can handle it, with which human checks and with data kept in the EU.
We will email you within one working day to book the 30 minutes. If it is urgent, message us on WhatsApp.
Prefer to talk right away?
Mon–Fri 9:00–12:30 and 14:00–20:00 (CET).