Local AI server with your own knowledge base
Your own AI assistant that knows your documents, answers with sources and runs entirely on your hardware.
AI assistants like ChatGPT are convenient, but send every question and document to someone else’s servers. A local AI server works differently: open language models run on your own hardware, and your data never leaves the network.
With a knowledge base built from your documents, notes and processes, it becomes a personal assistant. It answers questions with sources, summarizes, supports concepts and quotes and can later be extended with agents and automations.
BARNICOM plans the AI server, selects hardware and models, builds the knowledge base and maintains the system. On request, we connect it to your own cloud. More on AI without US providers: ChatGPT and Copilot alternative.
What a local AI server delivers
No data leaves
Language models run on your server. Documents, questions and answers never leave your network.
Answers from your knowledge
The AI searches your documentation, notes and processes and answers with sources instead of guessing.
Predictable costs
Hardware once, then no fees per user or request. Performance depends on the GPU, not on a pricing tier.
Building blocks
A typical setup of open components. Which models and hardware fit is defined in the concept phase.
| Area | Solution | Purpose |
|---|---|---|
| Hardware | Server with one or more GPUs | computing power for the language models, GPU memory determines model size |
| Models | open language models, for example from Mistral | understand, summarize and draft text |
| Model serving | Ollama or vLLM | provides the models within your network |
| Interface | Open WebUI or a custom application | chat for you and your team |
| Knowledge base | Retrieval-augmented generation (RAG) with a vector database | answers from your documents, with sources |
| Agents and automation | n8n or custom workflows | research, quotes, recurring tasks |
How the project works
-
Concept
Use cases, data sources, model choice, hardware and security are defined in a paid concept phase.
-
Hardware and models
Procurement and setup of the server, testing different models with your data.
-
Knowledge base
Your documents are ingested, structured and kept up to date, with access permissions preserved.
-
Expansion
Connection to your cloud, agents and automations, maintenance and model updates.
What to expect
Local AI is powerful but has limits. We tell you upfront where they are, for example:
- for very demanding tasks, the largest cloud models are sometimes still ahead
- larger models require expensive GPUs with plenty of memory
- answer quality depends on the quality and upkeep of your documents
Frequently asked questions
What hardware do I need?
A server with one or more GPUs. GPU memory determines how large the models can be and how many people can work at the same time. The selection follows your use cases.
What is RAG?
Retrieval-augmented generation: for each question, the AI finds matching passages in your documents and formulates the answer from them, with a reference to the source.
Does the AI really run without internet?
Yes. After the one-time model download, the server works fully offline. It only needs internet for updates, which you control.
Can the AI access my cloud?
Yes, approved folders of your own cloud such as Nextcloud. Both systems can also be built independently.
What does the switch cost?
It depends on the use cases, model size, number of users and hardware. You receive a proposal after the assessment.
From our projects
Self-hosting
Plan your own AI server
We discuss your use cases and propose hardware, models and setup.