Statistics and token usage
Statistics in the sidebar shows local conversation, request and token trends. Provider invoices and third-party account quota have different data sources.
Open
Select a period
Choose day, week, month, year or all history. Navigate periods or select a date, verify the displayed boundaries and refresh. Switch between compact and exact numbers when needed.
Interpret metrics
| Metric | Purpose | Caveat |
|---|---|---|
| Conversations | Count tasks | One conversation can make many requests |
| Requests / model requests | Observe call frequency | One user message can trigger multiple tool rounds |
| Input / output / total tokens | Inspect context and generation volume | Provider counts and local estimates can differ |
| User / assistant / tool records | Inspect record composition | Message counts are not HTTP request counts |
| Top models | Compare model usage | Check period and model IDs |
Trends and model rankings
Expand usage, request composition and model cards. Compare equal-length periods using the same number format. Multiplying total tokens by one price can be misleading because model, cache, reasoning and billing rules differ.
Example: a sudden increase
Select the affected period and inspect input/output proportions and model rankings. For input growth, check long histories, attachments and repeated reads. For request growth, use logs to investigate tool rounds, retries and background tasks. Statistics suggest where to look rather than proving the cause.
Data boundaries
Results depend on records retained on this device. Imports, deletions, missing history and absent provider usage affect totals. Provider statements determine charges. Check selected data before and after backup or migration.