---
title: Admin Assistant
description: Ask questions, jump to the right admin page, and make small site changes — menus and site settings — through an AI assistant built into the admin search modal.
section: guides
order: 11
updated: 2026-07-11
verified: 2026-07-11
related: [guides/ai-settings, guides/website-content, guides/ai-tools-and-provenance]
features: [admin-assistant]
---

# Admin Assistant

Get answers about your institution's admin tools — and make small, confirmed changes — without leaving the page you're on.

The assistant lives in the same search modal you already use. On any admin page, press **Ctrl+K** (or **⌘K** on a Mac) and switch to the **Assistant** tab. Editors and admins see the tab; other roles don't.

## Ask a question

Type a question the way you'd ask a colleague:

- "Where do I change the footer menu?"
- "What does the child-safe browsing policy do?"
- "How do I add a custom field to photographs?"

Answers are grounded in this documentation and in what the assistant knows about your current page, and they cite the doc pages they draw from. When the answer is "on another page," the assistant offers a **Go** button that takes you straight there — you'll only ever be offered pages your role can actually open.

The conversation survives navigating between admin pages while the modal stays in the session, but it's cleared when you close the modal or reload. Nothing you type is saved as a transcript.

## Let it make a change

For a small set of site-editing tasks, the assistant can do more than explain — it can propose the change itself:

- **Navigation menus** — create a menu; add, edit, reorder, or remove menu items (editor role or higher)
- **Site settings** — bind a menu to the public header or footer, and turn the site alert banner on or off (admin role only)

Ask in plain language: "Add a Visit link to the footer menu pointing to our visitor page."

The assistant never changes anything on its own. It responds with a **proposal card** describing exactly what would change, and the confirmation required scales with the risk:

- **Low-risk additions** (a new menu item) apply with a single click.
- **Edits to existing content** open a confirmation dialog summarizing the change.
- **Public-facing or destructive changes** (deleting a menu item, changing the public header or footer, publishing a site alert) show the exact before-and-after values and require you to tick an "I understand" acknowledgment before the confirm button activates.

You can always answer **Don't change anything** to dismiss a proposal.

Two safety behaviors to know about:

- **Stale proposals are rejected.** If the underlying menu or setting changed between the proposal and your confirmation — say a colleague edited it — the assistant refuses to apply and asks you to re-ask, rather than overwriting their work.
- **Unsaved edits block execution.** If the page you're on has unsaved changes, proposal buttons are disabled until you save or discard, so an assistant action never collides with a draft.

Everything the assistant applies is recorded in the audit log (**Admin → Audit Log**), attributed to you and marked as made via the assistant, with AI provenance like other AI-assisted changes ([AI Tools and Provenance](/docs/en-US/guides/ai-tools-and-provenance)).

The assistant deliberately cannot touch anything beyond the list above — no artifact metadata, no deletions of collections, no domain or safety settings, and nothing money-related.

## What it can see

The assistant works from your question, the conversation so far, this documentation, and a small structured summary of the page you're on (its title, and the names of items visible there). It never receives whole records, member emails, or credentials, and page content is treated strictly as data — instructions embedded in your catalog can't steer it.

## Cost and limits

Each question is one AI call billed to your institution's AI credits, visible in the AI usage view like any other AI feature. If your institution is out of credits, the assistant tells you and links to where credits are added.

Usage is rate-limited to keep costs predictable: 10 questions per minute per person, and 500 per day across the institution.

## Turn it off

The **Admin assistant** toggle in **Admin → Settings → AI & Cost Controls** disables the tab and blocks any pending proposals from being applied. Turning off **Enable AI features** disables it along with everything else ([AI Settings and House Style](/docs/en-US/guides/ai-settings)).