See your usage clearly
See tokens, models, and activity in one view.
Usage Stats reads the local ledger on this Mac and brings token totals, peak values, model usage, daily trends, and activity patterns into one view. The data stays local, and the 7-day and 30-day controls affect only the trend chart.
Open the statistics view
Expand the workspace sidebar, then select the pie-chart button in the top controls. Its tooltip reads Token Usage. Usage Stats opens as a full-window overlay and refreshes as new model turns are recorded.
- Start with the summary. Read total tokens, the peak day, longest chat, and both streak values.
- Compare model share. See which recorded models account for most of the tokens.
- Switch the trend. Use 7D for the latest movement or 30D for a broader pattern.
- Read the activity grid. Hover over a cell to see the date and its recorded token total.
The 7D and 30D controls change only Daily Token Trend. They do not rewrite the ledger or filter the summary and model totals.
Read the five summary metrics
The first row is a compact view of everything currently retained in the local ledger. Each number answers a different question.
| Metric | What it measures | Use it to understand |
|---|---|---|
| Total Tokens | Input increments and output tokens recorded in the retained ledger. | The overall volume of model work captured on this Mac. |
| Peak Tokens | The highest token total recorded on a single calendar day. | When the heaviest day of work occurred. |
| Longest Chat | The longest conversation span stored in the ledger. | How long the most extended recorded session ran. |
| Current Streak | The length of the most recent run of consecutive active days. | Your latest working rhythm, even if the run ended before today. |
| Longest Streak | The longest run of consecutive active days in the ledger. | Your most sustained period of recorded activity. |
These summaries report token usage, not session or message counts, and they are not a provider bill. They describe the usage events Levin recorded locally.
Read each panel in order
| Panel | Question it answers | Useful follow-up |
|---|---|---|
| Summary | What are the total, peak, chat span, and streak values in the ledger? | Use it as the baseline before reading the charts. |
| Model Usage | How are retained tokens distributed by recorded model? | Review the top five models; additional models are grouped as Other. |
| Daily Token Trend | How did model usage change across the last 7 or 30 days? | Switch ranges to separate a one-day spike from a broader pattern. |
| Token Activity | Which recent calendar days contain recorded tokens? | Hover over a cell to read its date and token total. |
All values come from the same local ledger. A chart is not a second source of truth.
Context composition
The context ring in the message composer shows how much of the active model window is occupied. After the first turn, hover over it to see the breakdown of messages, the system prompt, the Skills catalog, tools, and free space.
This ring belongs to the active conversation. It is a context view, not an additional quota or billing meter.
Know what the ledger keeps
The ledger keeps roughly 400 days of daily records on this Mac. Once usage is recorded in the ledger, archiving or deleting its conversation does not remove that history.
- Recorded: calendar date, model label, token increments, and the session span used for Longest Chat.
- Estimated: model totals derived from session statistics are marked with a tilde.
- Not a bill: provider pricing, credits, cached-token rules, and invoices are outside this view.
- Not a backup: the ledger belongs to this Mac and is not a copy of every conversation.
Accuracy and privacy
New usage is recorded when a model turn ends. For older conversations that predate the ledger, Levin can seed history only when reliable per-message usage is available. That is why these numbers can differ from a provider dashboard.
- Local source: every chart is computed from the usage ledger on this Mac.
- Model labels: usage keeps the model label captured when it was recorded; later configuration changes do not relabel history.
- Legacy gaps: an older conversation without reliable usage history is not reconstructed from guesses.
- Trend ranges: 7D and 30D change only the daily trend, not the stored records.
Use stats to make decisions
- Compare models with context. Pair token share with result quality and completion time; volume alone is not performance.
- Investigate spikes. Use the 7D trend for the immediate cause, then widen to 30D to see whether it is recurring.
- Notice the rhythm. Streaks and the activity grid reveal consistency without turning it into a target.
- Check the active conversation. Use the composer ring when the question is context pressure, not historical usage.
When a number looks wrong
First check whether Daily Token Trend is showing 7D or 30D; the other panels do not use that control. Then let the current model turn finish. If an old conversation never stored reliable per-message usage, Levin leaves it out instead of inventing a total. Remember that the ledger is local to this Mac and keeps the model label captured at record time.