Question 1

Hosting and GDPR compliance for the public sector?

Accepted Answer

Strictly EU hosting by default (Frankfurt). For Luxembourg municipalities, we can route via Mistral (FR) or Anthropic (EU) depending on sensitivity. Full audit logs of every citizen exchange, automatic anonymisation of personal data in logs, right to erasure. GDPR-compliant + Luxembourg national framework (CSSF where financial data is involved). For Dudelange: 100% EU hosting + Mistral for sensitive procedures.

Question 2

What happens during the first diagnostic call?

Accepted Answer

30 minutes by video call, no commitment. We talk about your business, your team, and the processes where things feel inefficient. By the end you leave with 2 or 3 concrete automation ideas and an indicative budget range. If we're not the right partner, we'll say so and point you in the right direction. No sales pitch during the call.

Question 3

How long does a project take to deliver?

Accepted Answer

Field audit: 3 days on site + 1 week to write the report. n8n automation system: 2 to 6 weeks depending on complexity. Conversational AI agent: 1 to 4 weeks depending on the number of channels and integrations. A working demo is typically delivered in the 1st or 2nd week so we can iterate.

Question 4

How much does an AI automation project cost?

Accepted Answer

n8n system: from 2,497 € excl. VAT for a small system (1 to 3 workflows), up to 4,897 € for a mid-size system, quote on request beyond that. AI agent: from 1,397 € excl. VAT for an essential agent, up to 2,347 € for an extended agent, quote on request beyond that. 3-day field audit: 2,500 € excl. VAT, deductible if you sign a quote within 90 days. Monthly maintenance: standard range 110 to 507 €/month for a system and 297 to 597 €/month for an agent. Beyond that (high volumes, multi-workflow, complex integrations, reinforced SLA), transparent quote on request.

Question 5

Are there hidden costs (LLM API, hosting)?

Accepted Answer

For AI agents managed on our EU infrastructure, tokens and hosting are included in the maintenance package up to the contracted volumes (250 or 500 conversations depending on the tier). Beyond that, +129 € excl. VAT per additional 500 conversations. For self-hosted n8n workflows, you pay your own infrastructure and third-party APIs directly (you know exactly what it costs, we are not an intermediary). For the European n8n cloud, budget 20 to 50 €/month depending on the plan, billed directly to you.

Question 6

Which AI models do you use (Claude, GPT, Mistral)?

Accepted Answer

Claude (Anthropic) by default for conversational quality. GPT-4o or GPT-5 for speed or ecosystem fit. Mistral (Paris) when strict EU data sovereignty is required (e.g. regulated sectors, accounting firms, public sector). The choice depends on your use case, budget, and sovereignty requirements. We can switch between models without rewriting the agent. On our Premium maintenance plan, we have already migrated agents from GPT-4 to Claude Sonnet in a matter of hours.

Question 7

Can you integrate my existing tools (CRM, ERP, accounting software)?

Accepted Answer

Yes. n8n has over 400 native integrations (HubSpot, Salesforce, Pipedrive, Airtable, Notion, Google Workspace, Microsoft 365, Slack, Teams, Shopify, WooCommerce, Stripe, Zendesk, Pennylane, Sage, Odoo, and more). If a native integration doesn't exist, we build it via custom REST API. Typical case: a client using a local Belgian ERP not supported by n8n, for which we wrote a dedicated connector in 2 days.

Question 8

What happens if the AI agent makes a mistake or hallucinates?

Accepted Answer

Three anti-hallucination layers: (1) RAG on your internal documents, the agent only answers from your content, not from its general memory; (2) strict system prompts that force the agent to say 'I don't know, I'll transfer to a human' rather than inventing; (3) post-processing safeguards (regex, classifiers) that filter questionable responses before sending. When the agent doesn't know, native human handoff to email, a Slack/Teams ticket, or a chat with full context. The Premium maintenance plan adds quality monitoring that flags unusual conversations for review.

Question 9

Does my data leave the EU?

Accepted Answer

By default, no. Strictly EU hosting (Frankfurt or Luxembourg for managed AI agents). n8n workflows hosted on your own infrastructure or on the European n8n cloud. For LLMs, Anthropic and OpenAI have European data centres and are covered by Standard Contractual Clauses for any potential transfers. Mistral is 100% European (Paris). If you have a strict legal requirement (CSSF, banking secrecy, public sector), we can route exclusively through Mistral to never leave the EU.

Question 10

Do I own my automations and AI agents?

Accepted Answer

Yes. You have full control over your automations and AI agents throughout the contract. The workflow code, system prompts, knowledge base, configurations, and documentation all belong to you. At the end of the contract (with notice), everything is transferred cleanly: if Apophisia was hosting on our managed EU infrastructure, we switch to your own cloud account or infrastructure; if we were using an external tool where you didn't have direct access (e.g. a shared platform account), we hand over full control with an explanatory guide for you or your next provider. No technical lock-in.

Question 11

What happens if I want to cancel my maintenance contract?

Accepted Answer

Maintenance contracts are signed for 3, 6, or 12 months depending on what we agree during scoping. You can choose not to renew at the end of the term by giving us 1 month's notice before the renewal date. Why no mid-term cancellation? Because we commit resources and optimise over time, and an AI agent improves over 6 months, not 4. In return, we continuously adjust to make sure results are there: if something isn't working, we talk and adapt the maintenance, we never leave you with a system that isn't running. During the notice month, we handle a clean handover: full archive, account transfers, and an explanatory guide for what comes next. Everything done properly.

Question 12

How do you train my team to use the system?

Accepted Answer

A 1 to 2-hour training session is always included in the delivery, by video call or on site. We document every workflow and every agent in a clear operator guide (PDF + Notion). For n8n workflows, your team can make minor changes themselves via the visual interface. For AI agents, prompt adjustments are documented step by step. If you want to build internal capability to manage everything in-house, an in-depth 2-day training is available on request.

Question 13

Can we start small?

Accepted Answer

Yes, and it's actually recommended. Many clients start with a single workflow or a single-channel agent to measure ROI before expanding. Recent example: an e-commerce client started by automating only product returns (2,497 € build cost), measured 12 hours saved per week over 3 months, then expanded to lead qualification and tier-1 support. The 30-minute diagnostic is designed exactly to identify the first brick that works for you.

AI agent for municipal administrations and public services