Metrics and insights
Charts
How to read every chart in Kairos — This Mac, the accounts' limits, the active account, tokens, API calls, cache, helpers, a session's or an account's resources, and the model picker.
Kairos draws its charts from what it keeps on the Mac: each account's limits at every check, Claude Code's own record of its requests, and a measurement of each session's CPU, memory, disk and network. Every chart says what it covers in its heading ("last 24 hours"), and every chart can be read aloud by VoiceOver.
Common to every chart
- Point at a time (or touch and drag on a phone) to read the values at that moment.
- All models picks one model for the token and API-call charts, or all of them together.
- Who used it last opens the sessions that used the most at the time you picked.
- Ask about this (the small button on a chart) opens Ask about that chart.
- "100% CPU is one core": a session at 250% is using two and a half cores.
This Mac
The CPU and memory of every session together, over the last 24 hours on Today, with Who used it (who, memory, CPU) at the time picked. Hardware draws the whole Mac, with its GPU, disk and network too, over an hour, a day or a week.
- "No session running. Kairos measures this Mac while one runs."
- "100% CPU is one core. Gaps: no session running, or the Mac asleep."
Accounts
Each account's 5-hour window, its weekly window and any other limit, on one 0–100% scale: "Each 0–100% of its limit". The dashed line is the limit ceiling. History draws the same over longer ranges, one row per account.
Active
Which account new sessions started on, hour by hour. Tap a time to see it.
Tokens
"All tokens each half hour: input, output and cache", over the last 24 hours.
| Part | What it shows |
|---|---|
| Tokens by model | How the tokens split between models. |
| Output by model, Output by account | Ranked bars: which model and which account wrote the most. |
| Cache hit | "Cache hit: the share of what a request sent that came from the cache." Higher is cheaper and faster. |
| Cache reads by context size | How much was read from the cache by requests of each context size. Large contexts re-read a lot on every request. |
| Helpers that read the most | The subagents whose requests read the most. |
"Counts your Claude Code sessions. Kairos's own runs are not included." With no tokens: "No tokens in this range."
API calls
The requests your sessions made, over the last 24 hours, in panels:
| Panel | What it shows |
|---|---|
| API calls by model | Calls each half hour, by model. |
| Helpers running | How many subagents were running. |
| Context per call | How large each call's context was. |
| Cache written | What calls wrote to the cache. "No cache was written by a call 5 minutes to an hour after the one before it." means no call paid to rebuild a cache that had just expired. |
Resources
A session's Resources (in its detail) and an account's Resources… (in its menu: "Every session on this account, added up") draw CPU, memory, disk writes and network over an hour, a day or a week, with Processes: what ran inside it.
- "Nothing measured yet. Kairos measures each session once a minute, and every few seconds while a chart of the last hour is open."
- "Mostly idle in the last day: under 5% of one core on average."
Week ahead and History
The forecast by day is explained on Week ahead; the limit rows on History. GitHub Actions, Repositories and Hardware have their own pages: GitHub Actions, Repositories, Hardware.
When a chart is missing
A chart that needs a newer engine says so: "Resource charts need Kairos 2.0.12 or later on the Mac.", "Kairos on this Mac does not count tokens yet; update it to see them." See Updates.