Private self-hosted AI stack

A private AI setup on self-managed infrastructure. Local models by default, with a cloud API used only where local hardware isn't enough.

The objective

Local language models served through Ollama behind an OpenWebUI front end, with SearXNG for private search. Anthropic's Claude API is wired in for the work local hardware can't do well.

What I did

Everything runs on infrastructure I control. Prompts and documents stay on the server unless a request is deliberately routed out to the cloud API.

The split is set per workload rather than per user, so it's clear which work leaves the machine and which doesn't.

The result

A working pattern for businesses that want to use AI without handing their data to a third party.