Ollama

Ollama runs language models locally. This shows the models and which are loaded, and the assistant can unload one to free memory.

Set it up

integrations:
  ollama:
    type: ollama
    url: http://ollama:11434

layout:
  - title: Infrastructure
    widgets:
      - ollama.models

Option	Type	Default	What it does
`url`	URL	required	Where the service answers, from Valet’s side.
`href`	string	optional	Where the card links to, when that differs from `url`.

One widget. Its key is ollama.models, using the instance name from your config.

ollama.models · list card · refreshes every 30 seconds while the page is open.

What the assistant can use, once its access level allows it. Read tools run freely. Everything else asks first.

Tool	Access	What it does
`ollama_unload`	control	Unload a model from Ollama’s memory, freeing its VRAM.
`ollama_models`	read	Models installed in Ollama, their size, and which are loaded in memory now.

What the page’s own buttons and controls can do. They need a signed-in person or a trusted network.

Action	What it does
`unload`	Unload a model from memory