temp-chart

A Linux terminal application that samples package-level CPU temperatures every 2,000 ms and plots both a 96-hour history and a six-hour detail view of:

• temperature readings in orange;
• rolling mean + 2 sigma and rolling mean - 2 sigma boundaries in cyan and blue.

The mean and population standard deviation use each sensor's trailing two-minute window, including the current reading.

The UI uses Ratatui. Samples are appended to a JSON Lines file and flushed after every batch. Restart downtime therefore appears as a gap instead of erasing history. The file is compacted to the 96-hour retention window at startup and every six hours.

Older data files without a persisted rolling mean remain supported; their means are reconstructed from the retained temperature history when loaded.

Chart data is incrementally aggregated into fixed-time min/max buckets. Normal redraws therefore process a bounded number of points based on terminal width, rather than rescanning the full 96-hour sample history. A full aggregation pass is only needed on initial display or after the terminal width changes.

Run

cargo run --release

By default, data is stored in ./cpu-temperature-history.jsonl. To put it elsewhere:

cargo run --release -- --data-file ~/.local/share/temp-chart/history.jsonl

List the Linux sysfs sensors the program selected:

cargo run --release -- --list-sensors

Press q, Esc, or Ctrl-C to quit.

Sensor selection

The collector reads Linux hwmon CPU drivers such as coretemp, k10temp, zenpower, and peci_cputemp. When package-level labels such as Package id 0, Tctl, or Tdie exist, core-level channels are omitted to keep the chart readable. If no supported hwmon device is found, CPU-related thermal zones are used as a fallback.

The process needs read access to the relevant files below /sys/class/hwmon or /sys/class/thermal.