Running with no AI

Valet is a homepage first. The assistant is optional, and leaving it out takes nothing from the page.

config.yaml

# no assistant: block
title: Home
integrations:
  downloads:
    type: qbittorrent
    url: http://qbittorrent:8080
    username: ${QBIT_USER}
    password: ${QBIT_PASS}
layout:
  - widgets: [downloads.active]

That’s the whole setup. There’s no switch to turn off.

What stays

• Every card, widget and integration.
• Every button and control, when sign-in allows them.
• The palette, for finding services and running actions.
• Live rings, notices, and notifications from live.notify.
• The year in review. No model writes it.
• History graphs on tiles.

The AI SDK and the MCP client load on the first question. With no assistant, they never load, so they cost no memory.

What goes

• The chat, and its button on the page. The ask box becomes a search box.
• The summary when you hover the status sentence.
• Tools, plans, reminders and pins, which all live in the chat.
• MCP, which only exists for the assistant.

hero.search: false removes the box as well.

In between

You can keep the chat and turn off everything it does on its own:

config.yaml

assistant:
  baseURL: http://ollama:11434/v1
  model: qwen3:8b
  background: { watch: false, brief: false, suggest: false }
chat:
  suggestions: fixed    # your list, with no model calls
  overview: false       # no ask box on the overview

Now the model runs only when you ask it something. With read access, the default, it still can’t change anything.