Careswitch has a built-in AI you can talk to in plain English to look things up and get everyday work done — finding clients and employees, checking shifts, reviewing invoices, and (when you allow it) making changes for you. The AI is branded Looper, and on the web it shows up in two related but distinct places: Command Center and the Ask Assistant side panel. This guide explains where to find each one, how they differ, how to keep your data safe, and how to phrase requests so you get accurate, useful answers the first time.

Note: Caregivers also have their own Looper chat in the mobile app. That mobile chat is separate from the Command Center and Assistant described here, which are web tools for office staff.

Both are powered by Looper, but they behave differently. Knowing which one you're in matters, especially before you let the AI change anything.

The simplest way to think about it: use Command Center for real work you want to keep and revisit, and Ask Assistant for a fast question about the record in front of you.

Command Center and Ask Assistant are workspace-admin only. If your role isn't a workspace admin, you won't see the Command Center item in the sidebar or the Ask Assistant action on detail pages at all, and visiting the page directly returns a permission error. This isn't a granular, per-feature setting — it's tied to being an active workspace admin. If you believe you should have access, ask another workspace admin to review your role under your workspace's team settings.

Either way, the AI can never do something you couldn't do yourself in Careswitch. Every change it makes re-checks your permissions at the moment it runs, so if your role can't manage billing, you can't have the AI create an invoice — even with Read-only off.

Command Center:

Ask Assistant:

Before the AI does anything, decide whether it's allowed to change your data or only look things up. In the message area you'll see the Read-only toggle (a shield icon; it shows the Read-only label and turns amber when on).

Important — there is no undo. Command Center does not ask you to approve or preview each action. The moment the AI decides to use a change tool, that change is committed to your live data immediately — no confirmation dialog, no draft step, and no rollback. If you tell it to cancel shifts or mark something completed and Read-only is off, it's done. This is exactly why Read-only matters: preview in read-only first, confirm the AI found the right records, then run the change.

In Command Center, your Read-only choice locks in for the whole session as soon as the session starts. After that the toggle is disabled, and you'll see a tooltip such as: "This session is read-only. The AI can look things up and answer questions, but it can't add, edit, or delete anything. Start a new session if you want it to make changes." When Read-only is off and locked, it reads: "This session is not read-only, so the AI can add, edit, and delete things. To use read-only, start a new session." So pick the setting before you send your first message.

If you started a session with the wrong setting, you can't flip it mid-conversation. The fix is simple and self-serve: start a new session from the Command Center page (or the new-session control in the sidebar) with the setting you want.

Because the Assistant side panel isn't a saved session, its Read-only setting is not locked — you can switch it on or off between turns in the same panel. It applies to your next message.

Tip: You can attach a file to your message if you want the AI to read something specific. This works in both Command Center and the Assistant.

The better your request, the better the answer. A few habits make a big difference.

The single most helpful thing you can do is say whether you mean a client (a care recipient) or an employee (a caregiver or staff member). Names and details overlap, so being explicit prevents mix-ups.

Include full names, dates, and date ranges whenever they matter. The more you pin down, the less the AI has to guess.

You don't need long explanations. State plainly what you want.

If you want more than one thing done, ask for each step in order. This makes the result easier to follow and check — and, with no undo, lets you confirm each step before the next.

Notice the difference: the strong version says it's an employee, gives a full name, and includes a clear date range. That's the pattern to aim for every time.

You don't have to start over. The fix is almost always to add the detail the AI was missing.

Start a Command Center session with Read-only on, and ask the AI to find the records and describe what it would change ("List the shifts for employee Emily Clark for the week of September 16, 2024 that aren't marked completed"). Once you've confirmed it found the right records, start a new session with Read-only off and run the change. Because there's no undo, this preview-first habit is the safest way to work.

Say you ask it to "Mark the shift for employee Emily Clark on September 18, 2024 as Completed." With Read-only off, the AI updates the shift immediately — no approval prompt — and it can't be undone in that session. That's expected behavior; just be sure the request named the right person, date, and surface before you send it.

The AI will tell you it can only look things up. In Read-only mode the change tools aren't available to it at all, so it can't make the edit even if you ask. To proceed, start a new session with Read-only off.

If you open Ask Assistant on a shift, ask a few questions, then close the panel or navigate to a different page, the conversation and the shift context are discarded. When you reopen the panel elsewhere, restate what you're asking about — or use Command Center if you want a conversation you can return to.

If a teammate doesn't see Command Center in the sidebar or Ask Assistant on a shift, that's the workspace-admin gate. A workspace admin can review their role in your workspace's team settings.

Command Center conversations are saved as sessions, so you can come back to them.

Command Center can run on a schedule so a routine task happens automatically — for example, a daily operations brief. From the Command Center sidebar, use the Schedule run control to set up a recurring run with a title, the prompt you want it to run, a time zone, and a recurrence. It then runs on its own at the scheduled times and the results appear as Command Center sessions. The same no-undo rules apply, so write scheduled prompts with that in mind.

Both are Looper, but Command Center is the full-page AI in the sidebar with saved sessions you can reopen, schedule, and continue — and its Read-only choice locks once the session starts. The Assistant is a quick, in-context side panel (opened with Ask Assistant on pages like a shift) that isn't saved; close it and the conversation is gone, and you can toggle Read-only on any turn.

No. Command Center commits changes the moment it uses a change tool — there's no approval step and no undo. That's why you should preview in Read-only first, confirm it found the right records, then run the change in a read-write session.

The setting locks in when the session starts so the whole conversation stays consistent. If you need the other mode, start a new session — that's the only way to switch.

They're available to workspace admins only. If your role isn't a workspace admin, neither one appears. A workspace admin can review your role in your workspace's team settings.

No. It's limited to your own permissions, which are re-checked when each change runs. And in Read-only mode, all change tools are removed entirely, so it can only look things up.

No. The side panel isn't saved. Only Command Center sessions persist — use Command Center if you want a conversation you can return to.

Yes. From the Command Center sidebar you can create a recurring run with a title, prompt, time zone, and schedule, and it runs automatically (for example, a daily brief).

Yes — both Command Center and the Assistant let you attach a file to your message so the AI can look at it.