Ollama · Open WebUI
Ollama · Open WebUI: a first setup, important settings, verification, and recovery considerations.
Working directory: ~/homelab/apps/ai. Follow the common setup workflow first. This is a setup note based on the linked documentation, not a copy of my live configuration or a runtime verification of this full stack.
Start here
Start with Ollama for model execution and Open WebUI for the browser interface. Use the upstream Docker example, choose a released image, and begin with a small model that fits your machine. A GPU is optional for this first test; do not add device permissions until CPU operation works.
Settings and data
Keep the Ollama API on the application network. Point the UI's OLLAMA_BASE_URL at the Ollama service name and port 11434, not localhost. Persist model storage and WebUI's /app/backend/data. Generate and retain WEBUI_SECRET_KEY; keep account signup restricted after creating your administrator.
Check it
Ask a harmless question and confirm which model answered. Restart the UI and verify that the test conversation remains. Observe RAM and disk use during a model download. Local model execution does not mean every configured integration stays local.
When something fails
If no models appear, check the provider URL, shared network, and Ollama logs. If generation is killed, try a smaller model before increasing limits. Back up UI data and its key; model downloads are replaceable, private conversations may not be.