Chatbot on its own server — German cloud or on-premises.
DigElite chatbots run wherever the customer wants them: on their own web server, in a German cloud with Hetzner, IONOS, STACKIT, or OVH Frankfurt, or as an on-premises installation behind the firewall. All three paths keep query data within the EU. We install, harden, and update the chatbot stack on the customer's hosting—no DigElite middle server, no external telemetry, and no data processing by us if the customer hosts themselves.
Three hosting options — the architecture remains identical.
Path A — On the customer's web server
For chatbots using API-based LLM (Aleph Alpha, Mistral), your existing WordPress server is sufficient. Install the plugin, enter the API key, index the knowledge base, and you're done. No server upgrade is necessary. Suitable for service chat, association chatbots, and administrative information bots.
Path B — Own server in a German cloud
For higher traffic or local LLM (Llama, Mistral on-prem), we recommend a dedicated server in a German cloud: Hetzner (Falkenstein, Nuremberg, Helsinki — the latter within the EU), IONOS (Frankfurt, Berlin), STACKIT (Baden-Württemberg, Black Forest), OVH (Frankfurt). GPU machines starting at around €200–400 per month will reliably cover a mid-sized business chatbot running Llama-3-8B or Mistral-7B.
Path C — On-Premises behind the firewall
For government agencies, associations with sensitive member data, or industries with confidentiality requirements: a complete stack behind the customer's firewall, with no internet hop for requests. We provide setup instructions, hardening configuration, and update path—operation is handled by the customer's IT team or an authorized service provider.
Comparison by application.
| Use Case | Path A (customer web server) | Path B (German Cloud) | Path C (On-Prem) |
|---|---|---|---|
| Service chat SME website | ✓ ideal | ✓ with higher traffic | – overkill |
| Association/Club Bot | ✓ ideal | - not necessary | - not necessary |
| Municipal citizen chatbot | ✓ possible | ✓ ideal (scaling) | ✓ in accordance with government IT requirements |
| Industry with secrecy | - not sufficient | – borderline | ✓ ideal |
| Email auto-reply | ✓ ideal | ✓ for many mailboxes | ✓ at official mailbox |
Example: Medium-sized city administration — Llama-3-8B on Hetzner.
A medium-sized city administration needs a citizen chatbot for online services. The IT department's requirements: no US cloud hop, in-house control over model updates, and predictable costs. We opted for path B: a dedicated server at Hetzner Falkenstein with Llama-3-8B, RAG indexing via the FIM standard service descriptions, and monthly hosting costs in the low three-figure euro range. The chatbot answers the majority of citizen questions directly and forwards the rest to the citizen services office via a structured ticketing system.
„"Anyone who sends a chatbot to a US cloud has given away their decisive advantage. We offer three hosting paths — all three with full data sovereignty, all three maintainable, all three affordable.""
— Philipp Herrmann, founder of DigElite
What potential customers should ask before deployment.
What hardware do I need for a local LLM (Llama, Mistral on-premise)?
For Mistral-7B or Llama-3-8B, a server with a modern GPU (NVIDIA L4, RTX 4090, or cloud equivalent) is sufficient. Hetzner GPU-equipped machines start at around €200/month, while dedicated servers with consumer GPUs start at around €400/month. For API-based models (Aleph Alpha, Mistral La Plateforme), you only need a standard WordPress server.
Is my server located at Hetzner Helsinki legally different from one at Hetzner Falkenstein?
Both are located within the EU, and both can be operated in compliance with the GDPR. Finland is subject to the GDPR just like Germany. If customers explicitly request "in Germany"—for example, government agencies with location clauses—we select Falkenstein, Nuremberg, Frankfurt, or Berlin. We document the location transparently.
What does DigElite handle, and what remains with the customer?
DigElite installs, configures, and hardens the chatbot stack once and provides an update procedure. The customer or their hosting provider is responsible for ongoing server operation (monitoring, backups, OS updates). We can offer a maintenance contract separately—this is then a traditional service, not SaaS.
What happens in the event of a server failure?
With path A, the chat component reacts to server outages like any WordPress plugin—the website is offline anyway. For paths B/C, we recommend standard high-availability patterns (failover, load balancer, regular backups). Since no external provider is involved, there is no "SaaS provider outage, nothing you can do" risk.
Three clusters that together support the GDPR argument.
Each individual pillar answers a sub-question. Only all three together result in a truly GDPR-compliant AI chatbot.
Where you can continue reading.
This feature is part of the DigElite chatbot family — check it out. Product Overview or the thematically related clusters.
15 minutes is enough to get an impression.
We'll be live-chatting with our own chatbot on nordzypern.live and showing you how it responds to real documents, when it honestly says "I don't know," and how it hands the call off to a human. No sales pitch, no Slide 47.
Watch the chatbot live & get an initial consultation