Healthcare / IT
IT and customer support
A major Dutch healthcare organisation was moving thousands of employees onto a new core platform at once, and expected first-line support to spike for the duration. Hiring for a short rollout was not practical, and nothing could be sent to an external AI service from inside their infrastructure.
What was built
The pieces, and how they fit
All of it is configuration on a single instance.
A self-hosted instance inside the organisation's own infrastructure, holding the application, the documents and the conversation state
An LLM provider pointed at a locally hosted model through Ollama, so no prompt and no document leaves the network
A collection of approved instructions and troubleshooting procedures, granted to the agent read-only and searched on every question
An agent that identifies the right procedure from a plain-language description and walks the employee through it
A function that hands unresolved cases to second-line support with the context attached, so nobody repeats themselves
Central administration of permissions, agent behaviour and integrations, exportable as configuration
The result
The assistant was used around 2,500 times in its first three weeks, supporting more than 4,000 employees through the rollout.
Self-hosted deployment
Local models via Ollama
Configurable LLM providers
Collections and file search
AI agents
Python functions
Persistent state
Role-based access control
Get started