Private AI and hosting
Self-hosted language models, a reverse proxy and a repeatable way to stand up, test and document each client service.
Portals, compliance and AITool on board, 2026
The aim
Use AI where it helps without sending client data to anyone: local models for drafting and search, and hosting that a small organization can afford.
- 0client records sent to outside AI
What was built
Local models
Models served on my own hardware behind one compatible API.
Proxy and routing
One proxy in front of every preview and client site, with HTTPS.
Host tooling
Scripts to prepare, back up and document each service, with a written host record.
Built with
- Linux
- Docker
- Caddy
- llama-swap
- OpenAI compatible API
- BookStack
Front end, back end, data and documentation by Kyle Nguyen. Screens use sample data; names are withheld and people in photos are blurred.
Need something like this?
Tell me what you are working with. I will tell you what a first version could look like, and how quickly.