What was delivered
- A private chat assistant (Open WebUI) for staff and the IT team, reached over the company network through one HTTPS gateway.
- Ollama serving open chat models, plus an embedding model so documents can be searched by meaning.
- n8n for building AI workflows that plug into internal tools.
- PostgreSQL with pgvector for data and search, and an audit gateway that logs every request to the models.
- Monitoring with Prometheus and Grafana, GPU health included, and admin screens for the containers and the database.
- Everything on one Windows server, with Linux running inside it through WSL and each service in its own Docker container.
Why it matters
It was the first AI that ran entirely on our own hardware. People could try AI on real work without sending anything to an outside service, and this version became the base that later grew into the full private GenAI platform.