What the metrics dashboard shows
The AI usage dashboard shows how many sessions ran and how many tokens they spent, split into input and output, plus an estimated cost. The totals sit on top; below them the same figures are grouped twice β once by model and once by UTC day β so you can see that most of the cost comes from one model or that a spike landed on a single day.
A small generator (scripts/generate-metrics.js) reads the recorded session list, sums input and output tokens per model and per day, and multiplies them by a per-model price table to estimate cost; the page then loads that snapshot. Ten deepseek-v4-flash sessions of 1,000 input and 500 output tokens each total 15,000 tokens β 10,000 input and 5,000 output β and about 0.0045 USD at 0.15 / 0.60 USD per million tokens.
How the numbers are aggregated
- Each finished session adds one session, its input tokens and its output tokens.
- Tokens are summed per model and per UTC day, taken from the session's last update.
- A per-model price table converts tokens into an estimated USD cost; a model absent from the table is still counted, but priced at zero.
Frequently asked questions about the metrics
How fresh are the numbers?
The page shows the timestamp of the last generator run, so the totals are as current as that snapshot.
Why is the cost only an estimate?
It is tokens multiplied by a fixed per-model price table, not a bill; models missing from the table add to the totals but cost nothing.
How long is usage data kept?
The snapshot covers the recorded sessions; the separate MCP tool-call audit log behind usage figures is swept daily and keeps the last 30 days.
