Usage dashboard: value at scale¶

The Tutor, Search and Helpdesk plugins and the AI provider deliver the AI features, and they are free. The Usage Dashboard is the operations layer above them: it makes running the suite measurable, manageable and defensible.

On a small pilot you may not need it. On a real, growing installation it is what turns "we have an AI assistant" into "we run AI responsibly — we know what it costs, we know it works, and we can prove it."

This page is the business case. For the tabs, settings and data model, see the Usage dashboard reference.

What it solves — especially at scale¶

See what your AI costs¶

Every chat answer spends tokens on your language model. Across many courses and users that cost adds up, and without measurement you are flying blind. The Tokens tab reports token consumption per day, per plugin and per provider instance, so you can justify a budget, spot a runaway course or user early, and attribute cost to a faculty or department.

Know every knowledge base still works¶

At scale, a course tutor's knowledge base may fail to parse, or an assistant may be deleted in RAGflow — and the block quietly answers "nothing found". Nobody can check hundreds of block instances by hand. The Status tab verifies every reference with a traffic light, tells misconfigured apart from RAGflow unreachable, and lets you refresh one area at a time. You find the broken one before a student reports it.

Show adoption and return¶

The Usage tab shows request volume and success rate over time, broken down by feature, by course, by user group (trainers vs. students) and the top users. That is the evidence a rollout needs: who uses the AI, where it lands, and whether it is growing — so you can decide where to expand.

Cut support time¶

When something breaks, "the AI does not work" is not a diagnosis. The Errors tab groups failures by labelled cause (rate limited, query too long for the embedding model, RAGflow server error, …), the API calls log shows the exact request and response, and an optional per-feature debug capture records a bounded request/response while you investigate. Root cause in minutes, not a ticket tennis match.

Oversight without storing user content¶

The usage log holds metrics only — no message content. Add optional anonymisation, a daily retention/purge task, a full Privacy API provider and admin-only access, and you have real oversight that your data-protection officer can sign off. For universities, public bodies and any GDPR-bound institution, this is often the decisive point: you can supervise the AI without keeping what users typed.

Report to the institution¶

The Export tab downloads the usage log for any date range as CSV, XML or PDF, all views in one file. That feeds institutional reporting, a board-ready PDF and chargeback figures per course or department.

Why larger installations feel it most¶

At scale	Without the dashboard	With it
Many block instances and courses	knowledge bases fail silently	the Status traffic light finds them first
Many users	AI cost is unknown	token trends + top users
Institutional scrutiny (budget, privacy)	no evidence to show	export + anonymisation + metrics-only
A support desk	guesswork	error types + the API log
Several faculties	no way to attribute usage	by-course / by-group breakdowns

Honest limits¶

Token accounting is an indicator, not a billing meter. It counts chat only (search uses no tokens) and only chats over RAGflow's OpenAI-compatible endpoint; chats that use session memory return no token data and are not counted; counting starts at installation; and the figures reflect what RAGflow reports, without guarantee. Read the token figures as trends and anomaly signals, not an invoice.

Low risk to adopt¶

The dashboard is optional and reads the provider's usage through an independent sink, so the other plugins work fully without it and nothing breaks if it is absent. It stores no message content by default, and you can remove it at any time. Each installed feature adds its own status checks and analytics through an rfdsource_* sub-plugin, so the picture stays complete as the suite grows.

Note: This documentation was created with the help of AI. Spotted an error, an omission, or something unclear? Please report it in the ragcon-docs GitHub repository so we can fix it.