Apps / Productivity
Local AI
A model that runs in your house
Ollama runs open-weight language models on this machine and Open WebUI puts a chat interface on them. Nothing you type is sent anywhere — which is the point for anything you would not paste into a website. Small models are usable on a CPU; anything large wants a GPU.
CategoryProductivity
Memory~2GB
Module
localaiInstalling it
Open the dashboard, find it in the App Store and press Install. Or from a terminal on the box:
sudo homebox install localaiWhere to open it
| Service | Port | First login |
|---|---|---|
| Open WebUI | :8181 | The first account you create becomes the admin. |
Open it at your box's address with that port — the dashboard links to it directly once it is running.
Good to know
- On 8181, not the 8090 the source uses — Beszel answers there. Two apps cannot share a port
- Start with a 3B or 7B model. On CPU those reply at reading speed; a 70B on CPU produces about a word every few seconds
- Models are large — a few gigabytes each — and live in the module's config directory on the system disk. Check you have room before pulling three
- The first account you create in Open WebUI becomes the admin, so make it before anything else on the network reaches the page