How to use the AI Assistant?

The AI Assistant opens on top of your current page — you stay in the section you are working in and ask questions in that context.

When you need this

You need a hint about the current section or where to find a feature.
You want to change settings by describing what you need (if your role allows it).
A multi-step task — review a plan first, then confirm.
Step-by-step navigation in the UI — a UI tour.
You need current information from the internet or data from a public website — if an admin enabled web search / URL fetch (see dedicated guide).

Good to know

The assistant button is the spark icon in the top bar (when enabled for your instance and role).
The panel title is AI Assistant.
Chat history is stored locally in the browser — it may not appear on another device or after clearing site data.

Before you start

An administrator enabled the AI Assistant on the instance.
Your role has access to the AI Assistant module.

Step by step

1. Open the AI Assistant

Click the spark icon in the top bar.
The AI Assistant panel opens on the left or right.
Empty state shows How can I help?; otherwise you see previous messages.

Tip: dock left/right or use a floating window — buttons in the panel header.

2. Choose a mode (optional)

Open Mode below the input.
Choose:
- Ask — answers only, no system changes;
- Agent — normal work with actions;
- Plan — plan first, then confirmation.

3. Send a request

Type your question or task in Ask a question about your platform....
Optionally attach a file (paperclip) if the file type is allowed in operator panel file settings.
Send or press Enter.

While processing, a status appears (e.g. Searching documentation..., Generating response...). Use Stop to cancel.

4. Use an agent skill (optional)

Type / in the input.
In the skill list (Skills in the UI), find the area you need (e.g. settings, clients).
Select a skill — a chip with its name appears.
Add your request text and send.

Remove a skill with × on the chip (UI hint: Remove skill).

5. Take a UI tour (if offered)

When the assistant offers to show the path in the UI, the chat shows Show me / No thanks.
Click Show me — step hints with highlights appear on screen.
Follow the steps or close the tour when done.

6. Approve a plan (Plan mode)

After your request, review Execution plan — steps and risk level.
Approve — the assistant runs the steps.
Revise — describe what to change; an updated plan appears.
Cancel — no changes are made.

7. Confirm a change

When the assistant in Agent mode wants to change data on the instance, it does not write it right away — a confirmation card appears in the chat.

Read the card: what exactly changes.
Click:
- Only this time — save this change;
- For this chat, for actions like this — save it and do not ask again for the same kind of action in this chat (not offered for risky changes such as deletion);
- Reject — save nothing.
The result appears under the card: Applied, Rejected or Could not apply (the buttons stay — you can try again).

Example: an AI QA criterion change. Ask: “In checklist 7, change the check instruction of criterion 11 to ‘The operator greets and says their name.’”. The AI QA criterion change card shows:

the path: Quality Assurance → Checklists → “Demo QA” (#7) → criterion “Greeting” (#11) → Check Instruction;
What changes: removed text is marked −, added text +, with some text around it;
the text length before and after (for example, “19 → 39 chars”);
Before and After — the full text with Show full and Copy;
when the change takes effect (for example, “Takes effect from the next assessment with this checklist.”).

AI QA criteria and settings offer only Only this time and Reject — each change is confirmed on its own. This card is shown to users who can make the change: for a criterion — with the right to save checklists, for AI QA settings — root only.

The 8 AI QA settings in Settings → Instance settings work the same way (for example, the checklist system prompt and the connection-check timeout). If a value is not set yet, Before shows “not set — default applies: …”. The assistant does not offer values outside the allowed range — it names the allowed limits instead.

If you have no permission. Without the right to make the change the assistant shows no card and saves nothing — it answers with text:

what it was going to change: the path, Before and After;
which right is missing: for a criterion — saving checklists, for AI QA settings — root only;
how to do it manually: Quality Assurance → Checklists (criterion) or Settings → Instance settings (key, root only).

Copy the path and the text and pass them to an administrator. The assistant answers the same way in read-only mode. If the right is removed after the card is shown, the change is not saved: the card shows “You have no permission to save checklists…” (for a setting — “Only a root user can change instance settings…”) and the save buttons disappear.

If the card says “The value changed after this card was shown” or “This card has expired”, nothing was saved — ask the assistant to prepare the change again.

8. Chat history

Open Chat history in the panel header.
New chat — a separate conversation topic.
Rename or delete via the menu next to the chat name.

After you finish

The answer appears in the chat; for complex actions the assistant may refresh the page so you see updates.
Verify important results — the assistant uses documentation and can be wrong on edge cases.

How to verify

The answer matches your request; changes (if any) are visible after a page refresh.
In Plan mode, changes happen only after Approve.
A change with a confirmation card is saved only when the card shows Applied.