WZ-IT designs, builds and operates production AI systems: internal assistants, RAG pipelines, AI features, LLM gateways and local inference on owned or European infrastructure.
Make company knowledge usable with sources and permissions.
Integrate AI features, agents and automation into existing systems.
GPU infrastructure, monitoring and operations from one team.
Leading companies worldwide trust WZ-IT
Four patterns we see again and again - and that one more tool rarely solves.
Prototypes impress, but never make it into secure day-to-day operation.
Documents, wikis, tickets, databases and APIs are not usable as AI context.
Chatbots sit next to the processes instead of being embedded in portals, workflows and systems.
Roles, sources, logging, monitoring, cost control and updates are planned too late.
Production AI needs four layers designed together: application, knowledge, inference and operations. We build and operate them as a whole.
Assistants, agents, UX, workflows, portals and business apps.
RAG, data sources, embeddings, Qdrant, sources and permissions.
Ollama, vLLM, models, GPU sizing and LiteLLM as the gateway.
Langfuse, monitoring, evaluation, cost control, updates, security and SLA.
Depending on your starting point we begin with the software, the company knowledge or the infrastructure - the target stays one productive system.
For companies that want to bring AI into existing portals, dashboards, internal tools or workflows.
For companies that want to make documents, wikis, tickets, databases and APIs usable with LLMs.
For companies that want to run models under control - locally, hybrid or hosted in Europe.
Concrete project patterns at the intersection of AI, software and infrastructure - not demos.
Answers with citations from your own documents and systems - with roles and permissions.
Connect and prepare distributed knowledge and make it searchable with high answer quality.
Assistance, extraction or classification right where your processes already run.
A familiar chat interface with local or controlled models - without prompts ever leaving the company.
One central access point to multiple models with budgets, keys and fallbacks.
LLMs read, classify and draft inside workflows - with boundaries, logs and review.
Take over AI-generated MVPs, harden them and make them operable.
AI Cube or GPU servers with operations, evaluation and cost control.
We pick components by use case, data protection, load and operating model - not by hype. Open source provides control, but the stack only becomes productive through integration, permissions, monitoring and operations.
Local inference engine
Run and manage models locally and expose them through APIs.
High-throughput inference
Tensor parallelism and OpenAI-compatible APIs for many concurrent users.
Chat and admin interface
User interface for local models, teams, tools and document chat.
Enterprise RAG app
RAG system for documents, workspaces, agents and knowledge bases.
Multi-LLM gateway
Central API layer for routing, budgets, keys and provider fallbacks.
LLM observability
Tracing, prompt management, evaluations, cost and quality control.
Vector database
Semantic search and retrieval layer for production RAG systems.
Automation building block
Connect processes, APIs and LLMs into controlled workflows.
We do not plan AI operations in theory. WZ-IT runs cloud, open-source and software stacks in production; with merkaio additionally IoT and remote-site systems for real sites such as ABCO Water and nextGYM. That operations experience feeds into local and hybrid AI systems.
Not every use case needs dedicated GPUs. The relevant factors are data protection, latency, cost, model size and operations responsibility. We size the stack to the risk and workload.
Send us the use case. We will respond with a pragmatic view on architecture, hardware and operations.
24.05.2026
Anyone bringing AI into production business processes quickly faces an uncomfortable question: what is actually happening in there? Which prompt went to which model, why...
10.05.2026
Three unauthenticated API calls. No login, no exploit framework, no privilege escalation. Three POST requests to a default port, and the machine's memory is on...
05.05.2026
On August 2, 2026, the EU AI Act's requirements for high-risk AI systems become enforceable. At the same time, the EU Parliament voted in April...
24.11.2025
With GPT-OSS 120B, OpenAI released their first open-weight model since GPT-2 in August 2025 – and it's impressive. The model achieves near o4-mini performance but...
09.11.2025
More and more companies are considering running Large Language Models (LLMs) on their own hardware rather than via cloud APIs. The reasons for this are...
Whether a specific IT challenge or just an idea - we look forward to the exchange. In a brief conversation, we'll evaluate together if and how your project fits with WZ-IT.
Leading companies trust WZ-IT
Timo Wevelsiep & Robin Zins
Managing Directors of WZ-IT
