AI Copilot shows the assistant's health and its index.
Status bar
A dot and the driver name. Green means the last probe worked (cached for 60 seconds), amber unknown, red failed, with a reason: credentials rejected, provider unreachable, embedding model not available, or no provider configured. "Quota/h" is the hourly limit per feature (chat has its own). The tooltip's help text about an AI_DRIVER setting is out of date: installs use the package assigned from LogaTech.
A yellow warning appears when stored passages were embedded at a different width than the provider now returns; they cannot be found until you reindex.
Counters
CMS documents and chunks, CRM documents and chunks. They are install-wide, not filtered by who is looking.
Knowledge Base card
Reindex CMS and Reindex CRM re-embed every document and then remove indexed documents that no longer exist. They run inside one request, cost AI usage, show no progress and no confirmation, and need only the Use permission. When they finish only the counters change. Changes to pages and CRM records are also indexed automatically as they happen, but failures there are logged, not shown.
Test semantic search
Type 2 or more characters and press Search: up to 5 matches with title, source type, a match percentage and a snippet, for what your account may see.
AI writing length
How long AI-written text is, in characters: the Product description (default 1,000), the Product short description (default 300) and a Blog post draft (default 3,000). Pick Short, Medium or Long, or Custom for any length from 200 to 6,000, then Save. Under each one is a rough token cost per click in Persian and in English, plus the request itself: longer text uses more of your AI tokens. Everyone who can use the copilot sees the lengths; only someone with AI Copilot's Settings permission (Users and roles) can change them; administrators have it.
Longer text takes longer to write, sometimes a minute or more on a model that thinks before it writes. If the AI reaches its length limit before writing anything, it says so and you can try again.
Quick prompts and chat
Four prompt chips send immediately; Clear conversation appears once there are messages. Answers that used look-ups show "N lookups".
Pending change card
Appears only on this page, above the privacy line, with Confirm and Decline. Confirming needs the "Let the AI propose changes" permission and the install switch.