Usage and limits
The tokens your agents use, what they would cost at API prices, your plan limits and the prompt cache.
Click Usage in the sidebar. The page adds up what every agent used, by day, model and project.
What the page shows
The row at the top picks what is counted:
| Control | Choices |
|---|---|
| Metric | Tokens, Requests (Turns when counting Kettu only), Cost |
| Agent | All, or one agent, once more than one has usage |
| Count from | This computer or Kettu only, shown while Usage on this computer is on |
| Time range | Today, 7 days, 30 days, 90 days, A year, All time, or Custom |
Custom opens a date picker with presets such as This week and Last month.
Below it:
- The figures: tokens, estimated cost, tokens per request, what the cache saved, the daily average and the busiest day.
- A chart per day, week or month; Today shows hours.
- Activity: tokens per day over the last 53 weeks.
- Tokens by type: input, output, reasoning, cache read and cache write.
- By model and By project.
- Plan limits.
Cost is estimated at API prices, and the page names the date of the price list. Tokens of models without a price are left out of the cost, and the page says how many.
Where the numbers come from
From the moment you install it, Kettu records the tokens of every turn it runs. That is Kettu’s record.
To include what the agents used before Kettu and in a terminal, turn on Usage on this computer in Settings › Usage, or click Count usage on this computer on the empty page. Kettu then also reads the logs Claude Code, Codex and OpenCode keep on this computer. It reads only token counts, models and times, never what was said, and nothing leaves this computer. It reads the logs again every 10 minutes, and Sync now reads them at once.
Settings › Usage also has:
- Each agent’s logs, with Re-import… to read them again from the start.
- Show in the sidebar: what the end of the sidebar’s Usage row shows: Today’s tokens, Today’s estimated cost, Tightest plan limit or Nothing.
- Clear Kettu’s record: deletes the tokens Kettu recorded. The agents’ logs, and what was imported from them, stay.
Plan limits
When an agent reports the limits of your plan, as Claude Code and Codex do, Plan limits shows each account’s windows, such as 5-hour and Weekly, with how much is used and when it resets. Agents that report none are named below the panel. Kettu reads the limits again when you open the page.
Routines pause when an agent comes near one of these limits.
Prompt cache countdown
Between turns, the top bar of a session shows how long its prompt cache has left, such as Cache 4 min. Click it for the cache’s lifetime, the last request, and about how many tokens the next message reads again at full price if the cache expires.
In the cache’s last minute, or its last five for an hour-long cache, the countdown turns to a warning color and the session’s row in the sidebar gets a ring.
Claude Code’s own transcript gives the cache’s lifetime, so its countdown is exact. For other agents on Anthropic or OpenAI models, Kettu estimates it from the shortest time the provider documents, and the countdown says Cache about 4 min. When neither is known, the top bar reads Cache time not reported.