Admin panel → Analytics — usage, cost, and traffic for your knowledge base. The menu item (the "Management" section) is visible to those with the Editor or Admin role in the current space (the "Space analytics" permission) and to the organization Owner (the "All analytics" permission). In a space where your role is Viewer, there is no such item — for an organization Admin too. The Owner sees search history across the whole organization; an organization Admin — for the spaces where they have the Editor or Admin role (in corporate spaces — by default), plus entries without a space; everyone else — for the spaces where they have such a role. The section is split into five tabs: Dashboard, Search History, Tokens, Traffic, Feedback.
Period selection (Today / 7 days / 30 days / 90 days) is available on the Dashboard, Tokens, and Feedback tabs. Search History uses an arbitrary date range, and Traffic uses a fixed 30 days.
#Dashboard
Three charts for the selected period:
- Queries and quality — how many search queries users submitted during the period (via the search portal, widget, API, or AI search mode), and the average quality score of the answers produced.
- Quality distribution — a pie chart: a breakdown of the "Quality" metric from the first chart — how answers were distributed across quality levels.
- Answer sources — a pie chart: what the answers were built from (the types are in the table below).
The legend is interactive. Clicking a legend item under any chart turns that data series on or off in the chart — for example, to hide cache and view only generated answers.
#Answer types (source)
| Type | What it means |
|---|---|
| AI-generated (LLM) | A fresh answer generated by AI based on your articles. |
| Cache (CACHE_HIT) | The answer was served from cache — this or a similar question has been asked before and the answer was saved. It's returned instantly and doesn't spend tokens. |
| Technical answer (fallback) | The AI could not produce a quality answer from your articles: either no relevant information was found, or the AI service was temporarily unavailable. Instead of generating one, the system returned a prepared technical answer. |
| Error | The query could not be processed for a technical reason: the base is not yet indexed or indexing is in progress, an index error occurred, or the query was blocked by a filter (meaningless input). AI unavailability does not fall here — that case goes to "Technical answer". |
#Search History
A chronological list of all queries. Filters: search by question text, date range (from / to), language, answer type. The CSV button — export to a file; it is shown to those with the "Export analytics" permission (by default — the organization Owner and Admin).
Table columns:
| Column | What it shows |
|---|---|
| Question | The text of the user's query |
| Language | The query's language |
| Type | Answer type: AI-generated / CACHE_HIT / Technical / Error |
| Cache | HIT — answer from cache, MISS — generated anew |
| Quality | The answer's quality score in % (for cached answers — a dash) |
| Tokens | How many tokens the answer consumed (for cache — a dash) |
| Latency | The time to produce the answer, ms |
| Country | The query's country (by geolocation) |
| Date | When the query was submitted |
#Tokens
AI token consumption for the period. Six metrics: Total queries, Total tokens, Tokens (input), Tokens (output), Indexing tokens (embedding when indexing articles), Avg. per query.
What tokens are. A token is the unit AI models use to measure the amount of text (roughly ¾ of a word, or ~4 characters). LLM consumption is counted in tokens: for the input (your question + the article fragments passed to the model) and for the output (the generated answer). Indexing tokens are spent once — when an article is created or changed and converted into a vector representation (embedding) for search. Your plan includes a monthly token limit, and this tab shows exactly that consumption.
The "Token consumption" chart has three series: Tokens/day (bars), Cumulative total (line), and Plan limit (dashed). The legend is interactive — a series can be hidden by clicking. Below are two breakdowns:
- By result type — AI-generated / CACHE_HIT / Technical.
- By AI type — search, editor, generation, chat, system, indexing (embedding).
#Traffic
Data for 30 days. Three metrics: Total queries, Countries, Languages. Below are lists with shares:
- By country — distribution of queries by geography.
- By language — distribution by query language.
- By result type — AI-generated / Technical answer / CACHE_HIT.
- Popular queries — the most frequent questions for the period with the number of hits (for example, "how to create a new article · 4×").
#Feedback
Only the organization Owner and Admin see this tab. "Was this article helpful?" ratings from users for the period. Three metrics: Total responses, Positive (%), Negative. Below are a chart by day, a breakdown of the reasons for negative feedback, and a list of articles with negative feedback (the link leads to the article editor). Until there are any ratings, the section shows "No feedback yet".