Why host your own AI stack instead of SaaS?

Three reasons. One: data sovereignty — your prompts and business data stay under Dutch law, no training data for an American provider. Two: cost control — no per-token credits model that escalates unexpectedly. Three: avoiding lock-in — you can switch LLMs tomorrow without breaking your pipeline.

Which LLM can I use — cloud or local?

Both. Our stack works with cloud APIs (OpenAI, Anthropic, Mistral, Groq) and with local models via Ollama (Llama 3, Mistral, Qwen, etc.). Often you combine the two: local for sensitive documents, cloud for heavy reasoning tasks. You choose per workflow.

Not for n8n, LangFlow, Flowise, Dify and OpenWebUI — they orchestrate LLM calls but run on CPU. You do need GPU if you want to run local models via Ollama: then we recommend a container with a dedicated GPU. We offer NVIDIA A4000 and higher for dedicated AI runtimes.

What does a RAG setup look like on this stack?

Typically: LangFlow for the retrieval flow, Qdrant or pgvector as vector database, optional Ollama for embedding models, and n8n to periodically index documents from the source (SharePoint, Drive, helpdesk). Four containers, no external dependencies except your chosen LLM.

Is this stack GDPR-compliant?

The infrastructure is — two ISO 27001 data centres in the Netherlands, processor agreement on request. GDPR compliance of the end result depends on your use case: which data you process, which LLM you pick, how long you retain conversations. We help with the DPIA if you want.

Does this stack work with Azure OpenAI or AWS Bedrock?

Yes. n8n and Dify have native integrations with Azure OpenAI, AWS Bedrock and Google Vertex AI. Many customers use Azure OpenAI for compliance-driven workloads and local Ollama for experiments — the orchestration layer stays the same.

AI Infrastructure on your own environment. Privacy by design.

The building blocks for a privacy-friendly AI stack: orchestration, agent frameworks, vector databases, your own LLM interface and LLM runtime. Each component in its own Docker container, fully under Dutch law. No prompt leakage to OpenAI or Anthropic — unless you want it.

A complete AI stack — loosely coupled

No vendor lock-in. Combine n8n for workflows, LangFlow or Flowise for agent flows, Dify as an LLM-app platform and OpenWebUI as the chat interface on your own models. Each component standalone-usable, more powerful together.

GDPR-compliant — data stays in NL

Containers run in two ISO 27001-certified data centres in the Netherlands. Optional Ollama runtime keeps prompts fully inside the EU. Cloud LLMs (OpenAI, Anthropic, Mistral) are a choice, not a requirement.

Full control over the runtime

Your own API keys, your own models, your own vector database (PostgreSQL pgvector, Qdrant or Weaviate). Your own logging, your own telemetry. We manage the runtime — you keep the data and the policy.

Stack live in 5 minutes

Data stays in NL (GDPR)

Open-source, no lock-in

GPU runtime on request

The building blocks of your AI stack

A production AI application typically has 5 layers. We host the runtime; you design the stack.

Orchestration

n8n or Activepieces — connects LLMs, data systems and business processes. No Zapier data leakage, no credits meter.

n8n · Activepieces

Agent frameworks

Build agents visually with LangFlow or Flowise — drag & drop on top of LangChain. Ideal for RAG, multi-step reasoning and tool use.

LangFlow · Flowise · Dify

LLM interface

A ChatGPT-like interface on your own models. Per team, per project or per customer. Works with OpenAI, Claude, Mistral and local models via Ollama.

OpenWebUI

LLM-app platform

Build production LLM apps and knowledge bots with versioning, datasets and evaluation. Dify is a complete platform solution.

Dify

Optional runtime + vector DB

Your own Ollama runtime for local models (Llama, Mistral, Qwen) and a vector database for your RAG pipeline.

Ollama · pgvector · Qdrant

What you can build with this

Your own RAG over company documents

Search SharePoint, contracts or helpdesk history with LLMs without the source documents leaving your infrastructure. LangFlow + Ollama + Qdrant in three containers.

Customer-service AI with human escalation

n8n routes tickets, Dify drives the LLM conversation, Chatwoot handles live chat. One pipeline, three containers, you keep the transcript.

Private ChatGPT for your team

OpenWebUI as chat interface on top of your own API keys. Everyone gets access to GPT-4, Claude or a local model — without prompts leaking to third parties.

Automatic reports and summaries

n8n pulls data from your systems, an LLM summarises, n8n delivers the report via email or Teams. Not an 'AI feature' in a SaaS — your own pipeline.

Our AI components

Every component runs in its own Docker container with persistent volumes. Updates follow the official upstream track, your data stays yours. Combine as you like or ask for a complete-stack advice.

Most chosen