AI Agents for the Service Desk
Agents that sort, summarise and route tickets and draft answers, connected to Matrix42, Jira or ServiceNow through n8n.
AI agents and language models for your processes
We build AI into business processes: agents that work in the service desk, language models that run on your own hardware when data has to stay in-house, knowledge bases that answer with sources and models trained on your own data. Everything runs inside governed n8n workflows.
What we build
Local LLM Hosting
Language models on your own servers or in an EU data centre, served with vLLM, llama.cpp or Ollama and connected to your workflows through n8n.
RAG Knowledge Base
A chat that answers questions from your documents and shows where each answer comes from, built with n8n and a vector database.
LLM Fine-Tuning with LoRA
Language models adapted to your terms, formats and tasks with LoRA and QLoRA, at a fraction of the cost of full training.
Questions we often get
Yes. The agents run as n8n workflows. For Matrix42 they use our open-source Matrix42 node, which works with tickets, journal entries and data objects. Jira, ServiceNow and other systems connect through their APIs.
Only if you want it to. Many tasks run well on small open models with vLLM, llama.cpp or Ollama on your own server or in an EU data centre. Where a task needs a larger cloud model, we agree beforehand which data it may see.
Only as far as you allow. The agents suggest and people decide. Every change an agent makes is logged. Actions that cannot be undone need a confirmation.
RAG looks up the answer in your documents every time and shows the source, which suits knowledge that changes. Fine-tuning changes how a model writes and decides. It suits fixed formats, your own terms and narrow tasks such as classifying tickets. Most projects start with RAG.
That depends on the model. Models with 7 to 8 billion parameters run on a single workstation graphics card. Larger ones need more graphics memory or several cards. We size the hardware for your tasks before you buy anything.
With your data. We look at your tickets or documents together, pick the task that takes the most time and build one agent or knowledge base for it. It runs next to your team until it works. Then the next task follows.