For IT departments · HainzelStack

The AI platform that leaves you in control.

HainzelStack is the modular, LLM-agnostic platform behind our digital employees — EU-hosted, integrates in a standardised way and free of vendor lock-in. You keep full data sovereignty and technological flexibility.

Book a technical conversation → View architecture
EU-hosted On-prem possible ISO 27001 · GDPR
HAINZELSTACK · CONTROL EU · Frankfurt
Model per task
Claude GPT-4o · Azure Gemini Llama 3 · self-hosted
Active roles
Knowledge Assistanceactive
RFQ supportactive
Reporting Assistancerunning
Audit log · every action logged & traceable

Why HainzelStack

Sovereign. Modular.
No lock-in.

01

Full data sovereignty

All data, prompts, retrieval indexes and logs stay in the EU. Customer data is not used to train foundation models.

02

No vendor lock-in

LLM-agnostic and modular by design. Models, connectors and components can be swapped individually — you stay independent.

03

No IT-project risk

Standardised and introduced step by step. No monolithic migration — you start small and expand in a controlled way.

Architecture

HainzelStack — six layers, full control.

A cleanly cut layered architecture from EU hosting up to the application. Every layer is replaceable and individually auditable — the basis for data sovereignty, compliance and fast expansion.

Request an architecture deep-dive
06ApplicationApps, AI Studio, self-onboarding & admin interfaces.
05ExecutionTool calls, workflows, guardrails, human-in-the-loop.
04ModelingLLM routing, retrieval, embeddings, evaluation harness.
03Agent InteractionRoles, prompts, escalation & handover logic.
02MessagingOutlook, Teams, WhatsApp, web chat, email.
01TransportEU hosting, authentication, encryption, tenant separation.

Agent operating system

Hainzeloop: every role runs on it.

Every digital employee runs on Hainzeloop, our own agent operating system. The roles at our customers as well as the ones our own team works with. Not bought in and not assembled from parts: built by us, operated by us.

For you that means: you fill five roles and still operate one system. Whatever you have reviewed and approved once applies to every further role.

Does this also run in our data centre?

Yes. Hainzeloop is a single program that gets copied and started. No database, no third-party service, no runtime to install first. One server is enough, whether in the EU cloud or on your own premises.

What did the role do, and what did it cost?

Every step is recorded: which request, which model, which tool, how long, what it cost. Visible in a built-in dashboard. Not a black box but a file you can put in front of an auditor.

Does this tie us to one AI provider?

No. The language model can be swapped without changing anything about the roles. Every request starts on an inexpensive model and only steps up when the task calls for it.

Where do our data and credentials end up?

The data stays where Hainzeloop runs. Credentials live only in the server environment, never in configuration files and never in the shipped package.

How is a role tailored to us?

What a role should be able to do is written in plain readable text files, not in program code. Changes therefore need no new release. A new role is a new description, not new software.

Do the roles reach people where they work?

Teams, WhatsApp, Discord, the browser and an interface for your own systems. All channels share the same memory, so a case started on a phone can be continued at the desk.

Agents from the open-source toolkit

  • Every role is assembled separately and maintained separately
  • Several services have to work together: database, runtime, containers
  • Logging, cost overview and audit evidence are things you add yourself
  • When something breaks, the community helps if it has time

All roles on Hainzeloop

  • One runtime for every role, reviewed once instead of five times
  • One program, one file, one start
  • Logging, costs and evidence are there from the beginning
  • We run it for our own roles and we stand behind it

HainzelStack describes the platform from EU hosting to the application. Hainzeloop is the part inside it that actually runs the roles. Open-source projects show what is possible with your own agents. For a business that has to account for what happened, they usually lack exactly the parts that take effort: traceability, operations, and someone who is answerable.

LLM-agnostic

The right model for every task.

HainzelStack picks the suitable model per use case — by capability, cost and data sovereignty. Providers are interchangeable without you having to rebuild roles or integrations.

A

Anthropic Claude

For demanding reasoning and long contexts.

O

OpenAI / Azure

EU region via Azure, broad tool support.

G

Google Gemini

Multimodal and cost-efficient for standard tasks.

∞

Open source

Self-hosted (e.g. Llama) for maximum data sovereignty.

Integration

Fits into your system landscape.

Pre-configured connectors for the tools SMEs use — heterogeneous stacks included. Whatever is missing, we connect via standardised interfaces.

Microsoft 365
DATEV
SAP Business One
Sage
Lexware
Outlook & Teams
REST & webhooks
Your own ERP / CRM

OAuth-based connection · predefined templates for common data sources · custom connectors on request.

Compliance & operation

Security and verifiability, built in.

EU-hosted & on-prem

Operated in Germany (Frankfurt) — optionally entirely in your own data centre.

EU-AI-Act-compliant

Hainzelman acts as the provider. Risk classification and technical documentation per role.

Monitoring & audit trail

Every agent action is logged — visible on the customer side for GDPR and AI-Act evidence.

Human oversight

Defined escalation and handover points. The agent asks and hands over when a human must decide.

Funding

Funded by the State of Schleswig-Holstein.

The development of the Hainzelman Platform — a ready-to-use, secure reference implementation based on HainzelStack — is funded through the “AI Funding Programme” of the State of Schleswig-Holstein. Companies keep full data sovereignty and technological flexibility, avoid dependence on individual AI providers and can swap components later.

Contact

Talk to our team.

Architecture, integration, operation, compliance — we go through the details with your IT and plan a gradual, secure rollout.

Robin Müller

Managing Director
04101 — 8054255 [email protected]

Book a technical conversation

Architecture, integration & compliance — straight with the team.

or
Book a meeting directly