Statistics

The Statistics page (admin console nav label "Statistics") shows platform-wide usage aggregated across all users and tenants. It is the operator's view of how much the install is being used, what it costs, and which tools and models dominate — distinct from a single user's own usage view, which is scoped to the requesting user only.

All figures are read from the billing-grade usage_records table (tokens and cost) and the runs table (run counts, success/failure), so they match what the cost and budget subsystems charge against.

Time range

A single control in the top-right selects the window: 1 day, 7, 30 (default), 90, or 365 days. Changing it reloads every chart and tile.

Hourly buckets at 1 day

Selecting 1 day switches all time-series charts from daily buckets to 24 hourly buckets (axis labels become 21:00 instead of MM-DD).

KPI tiles

A row of summary tiles covers the selected window:

Tile	Meaning
Tokens	Total input + output tokens (abbreviated, e.g. 1.2M)
Cost	Total spend in USD
Runs	Distinct runs; sub-label shows the run success rate (% OK)
Active users	Distinct users with activity in the window
Avg cost/run	Cost divided by runs
Cache hits	Cache-read tokens; sub-label shows the cache hit rate
Failed	Failed runs (runs.status == failed)
Avg tokens/run	Total tokens / runs
Tool calls	Total tool calls; sub-label Ø n/Run
Tool failures	Failed tool calls; sub-label shows the failure percentage
Chats	Platform total chats (not windowed); sub-label Ø n/user

Note

Chats and its per-user average are all-time platform totals ("how big is the install"), unlike the other tiles which respect the time range.

Charts

• Active users & runs per day — a combined chart: successful vs. failed Runs as stacked bars, with Active users overlaid as a line on a second axis.
• Tokens / cost per day (by model) — stacked bar charts broken down per model; the heaviest models are stacked and the remainder folds into an Other series.
• Tokens by model — a doughnut of each model's share of total tokens over the range.
• Cache per day — cache-read (Cache hits) vs. cache-write tokens per bucket, with the overall hit rate in the header. Shows a hint when no prompt caching occurred in the range.
• Tool calls per day and Top tools — tool-call volume over time and the most- used tools, with failure counts (shared with the per-user view via the tool_stats service).

Per-user breakdown

Below the platform charts, a Per user section ranks the top 10 users by tokens (labelled by email, or a shortened subject if no email is known) as horizontal bar charts:

• Tokens per user (split into input / output)
• Cost per user (USD)
• Avg tokens per run by user
• Chats per user
• Tool calls per user

How it differs from a user's own usage

The per-user Usage page shows only the signed-in user's own tokens, cost, runs, budget and tool stats (server-side scoped to their subject via /usage/*). The admin Statistics page aggregates the whole platform across all tenants, adds active- user counts and per-user rankings, and is reachable only from the admin console.