Two projects, one template
Why self-host Ollama and Open WebUI
Ollama pulls open models and serves them behind a local API. One command gets Llama, Qwen, Gemma, DeepSeek or gpt-oss running. The endpoint on port 11434 speaks the OpenAI format, so most existing client code points at it with a change of URL. The project is MIT licensed.
Open WebUI is the chat interface that ships in the same template. It handles logins, document upload for RAG and per-group permissions. It also talks to any OpenAI-compatible backend, so Ollama doesn't have to be the only thing behind it. Running both on your own VPS keeps the prompts, the uploaded documents and the model weights on a disk you rent instead of a vendor's.