AI usage dashboard

Lives in apps/web (the admin console), not apps/lms - reachable by admin, superadmin, and support.

Who sees what

• A tenant admin sees only their own tenant's usage.
• Superadmin and support are treated as platform actors: they get a cross-tenant rollup (per-tenant totals) with a ?tenantId= link to drill into any one tenant's own breakdown - the same view a tenant admin gets for their tenant, just reachable for any tenant.

What it shows

• Usage broken out by instructor, by learner, by system (background jobs, e.g. knowledge-base ingestion, not tied to any one user), and by content type (lesson content, exam questions, blog posts, knowledge-base ingestion, course Q&A, and so on).
• A daily trend of input/output token counts over time.
• Each row's call count and token totals.

Reading the numbers

Most rows carry real token counts reported by the AI provider. Rows flagged estimated are the exception: Gemini's embedding API (used for knowledge-base ingestion) doesn't return a token count at all, so those rows are estimated from character count instead (roughly 4 characters per token) - a deliberate approximation, flagged so the dashboard never presents it as an exact metered number.