Operations & back office
Order intake, document chasing, status updates, and hand-offs between email, Excel, and your ERP.
New · AI Agents
AI agent development in Luxembourg
Not another chatbot
Chatbots talk. Agents get work done.
Most “AI” in business is a chat window that cannot touch your ERP, email, or files. An agent is different: it has a goal, tools, memory, and rules. It can draft, look up, update, and escalate, then stop and ask a human when the stakes are high.
- Wired into your operations
- Agents call the same APIs and apps your team already uses: ERP, CRM, email, spreadsheets, internal tools.
- Under your control
- Clear permissions, audit logs, and human-in-the-loop steps. No black-box surprises in production.
- Designed in Luxembourg
- Architecture, data handling, and support from a team that understands GDPR and how SMEs here actually work.
Where agents pay off
Built around the work you already do
We start from one painful, repetitive workflow, not a science project. Typical first agents in Luxembourg SMEs:
Customer support
Triage incoming requests, pull the right account context, draft replies, and escalate the exceptions.
Research & reporting
Collect sources, summarise changes, and produce a weekly brief your team can actually use.
Internal copilots
A specialist assistant for finance, HR, or project delivery that knows your policies and templates.
How it works
A controlled loop: observe, decide, act, verify
Every agent we ship is a system, not a prompt. Triggers start a run. The runtime chooses tools. Memory supplies context. Guardrails stop unsafe actions. You see what happened.
Trigger
Agent runtime
Tools
- ERP
- APIs
Memory (RAG)
Guardrails & human approval
EU hosting
Stack
The technologies we actually use
We pick the model and tooling for the job, not a one-vendor demo. Your data stays in systems you control, hosted in Europe.
Models
- OpenAI
- Anthropic Claude
- Open-weight models
Orchestration
- LangGraph
- LangChain
- Custom runtimes
Automation
- n8n
- Webhooks
- Queues
Memory & RAG
- RAG
- PostgreSQL + pgvector
- Document stores
Tools & integrations
- MCP
- REST & GraphQL
- ERP / CRM / email
Hosting & compliance
- EU cloud
- GDPR-aware design
- Audit logs
How we deliver
From discovery call to an agent in production
A short, honest path. If an agent is the wrong tool, we will say so on the first call.
Week 0
Discovery call
15 minutes. We pick one workflow, the systems it touches, and whether an agent (versus simpler automation) is worth building.
Week 1
Map the work
We shadow the process, list tools and failure cases, and define success in numbers you already track.
Weeks 2–3
Prototype the agent
A working loop on a narrow slice: observe → act → verify. You can click through real runs, not slides.
Weeks 3–5
Connect your tools
APIs, MCP, email, ERP, files. Least-privilege access. Logs for every tool call.
Week 5
Guardrails & evaluation
Approvals, rate limits, eval sets, and a clear “what the agent must never do” list.
Ongoing
Deploy & operate
EU hosting, monitoring, and a Luxembourg contact who can change the agent when your process changes.
Start with a discovery call
No deck. We look at one painful workflow and tell you whether an agent is the right move, or whether a simpler automation will do.
Book a discovery call
Pick a time that works for you. We'll talk through your project and see how NexClass can help.
FAQ
Questions we hear before the first agent
What is an AI agent, in practical terms?
Software that can pursue a goal across multiple steps: it reads context, chooses tools (email, ERP, APIs), acts, and checks the result. Unlike a chatbot, it is allowed to change something in your systems, within the rules you set.
How is this different from ChatGPT or a website chatbot?
Consumer chat tools answer questions in a box. Our agents are integrated into your stack, have permissioned tools, keep an audit trail, and can be required to get human approval before sending, paying, or updating records.
Do you train on our company data?
No. We use retrieval (RAG) and your existing systems so the agent can look things up. We do not send your data into a public training set. Hosting and vendors are chosen so data can stay in the EU.
Is this GDPR-aware?
Yes. We design for data minimisation, access control, logging, and EU hosting. A Luxembourg-based team remains the accountable contact for how the agent is built and run.
How long until we see something working?
A discovery call is 15 minutes. A first prototype on one workflow is typically a few weeks, not a six-month lab project. We would rather ship a narrow agent that runs than a broad demo that does not.
How do we start?
Book a discovery call. Bring one repetitive process (for example: “every order confirmation still needs a human to copy data into the ERP”). We will tell you if an agent is justified.