Skip to content

Releases: akivajp/rigtop

rigtop v0.9.0

Choose a tag to compare

@github-actions github-actions released this 26 Oct 00:51

Added

  • Graph time span: fit to the width (one sample per pixel or braille dot,
    as before), or the last 1, 5, 15 minutes or hour, averaged or stretched to
    the width; g in the TUI, the "graph" button in the GUI's CPU panel title
    (remembered between runs). The same span applies to every host. Graphs
    keep 4096 samples (over an hour at the default interval).
  • [alerts] in the configuration sets when readings become alerts:
    filesystem usage (90% warning, 97% critical), and new warnings for hot
    CPUs (96 °C), GPUs (85 °C) and drives (70 °C, from SMART or the NVMe
    sensor); 0 turns one off. Remote hosts are judged by the viewer's
    settings.
  • GUI: a "Sensors" window (the "sensors…" button under the CPU cores, or a
    click on a sensor's name) lists every temperature and fan channel with its
    reading, to name it or hide it; saved to [sensors] in the configuration
    file with the rest of the file left as it was.
  • Debian packages in the releases (amd64 with the GUI and a desktop entry,
    arm64 with the terminal UI), and a Scoop manifest for Windows
    (packaging/scoop/rigtop.json, adds a Start menu shortcut).
  • --demo (TUI and GUI): made-up data for three hosts (a workstation
    training on two GPUs, a laptop on battery, a NAS with alerts) to try the
    interface; every poll is a second of made-up time, so a short
    --interval-ms fast-forwards. The README screenshots use it.

Fixed

  • TUI: the CPU model no longer runs into the core grid next to the graph

rigtop v0.8.2

Choose a tag to compare

@github-actions github-actions released this 25 Oct 20:28

Added

  • GUI: a "widgets" button (next to "hosts…") hides and shows the widget
    column, remembered between runs

Fixed

  • GUI on narrow windows (e.g. with the widget column): the process table
    drops Command, Thr, User, VRAM and Gpu% in that order instead of hiding
    Cpu% and Mem off the edge; the CPU model no longer runs into the core
    grid, which also gets a denser cell for short panels; the CPU info rows,
    GPU names, drive rows and btrfs rows wrap instead of being cut off; link
    speeds move to a tooltip so network rates stay visible; the title clock
    hides when the controls need its place; the overview puts cards side by
    side when one column would leave them too short for all their sections

rigtop v0.8.1

Choose a tag to compare

@github-actions github-actions released this 25 Oct 20:01

Added

  • Widgets: "gauge_style": "level" draws a row's gauge in its level color,
    for amounts left (where a full bar is good)
  • contrib/widgets/claude-code/: usage windows shown as what is left
    (RIGTOP_CLAUDE_QUOTA=used for usage); idle sessions show for how long
    and whether they wait for you or for background tasks and subagents;
    session names without the project prefix

rigtop v0.8.0

Choose a tag to compare

@github-actions github-actions released this 25 Oct 19:41

Added

  • Widgets: a column on the right (GUI and TUI, in the overview and host
    views) with a built-in large clock and [[widgets]] fed by any command
    that prints text or JSON rows (labels, values, gauges, colors); w
    toggles it in the TUI
  • contrib/widgets/claude-code/: a widget showing Claude Code's usage
    limits (5-hour and weekly), running sessions and active subagents

rigtop v0.7.1

Choose a tag to compare

@github-actions github-actions released this 25 Oct 18:28

Fixed

  • Windows: NVMe drive temperatures from LibreHardwareMonitor ("Composite
    Temperature", or numbered sensors), and values it sends as text with
    units

rigtop v0.7.0

Choose a tag to compare

@github-actions github-actions released this 25 Oct 18:15

Added

  • GUI (Linux, macOS): desktop notifications when a new alert appears on
    any host, also while minimized ([notifications] level: critical by default,
    warning or off); alerts present at startup are not announced;
    rigtop-gui --test-notification shows a sample one
  • Windows: drive health from LibreHardwareMonitor (wear, spare capacity,
    temperature, power-on hours) in the SMART rows, as disk0, disk1...
    like Disk Management; drives without a disk row also show their
    temperature (and the model on hover in the GUI)

Changed

  • NVIDIA: per-process GPU utilization, the costliest NVML query (about
    2 ms per GPU), is skipped for GPUs no process uses and while only the
    overview's top processes are wanted

rigtop v0.6.0

Choose a tag to compare

@github-actions github-actions released this 25 Oct 17:11

Added

  • Linux: AMD and Intel GPUs in the GPU panel and on overview cards (usage,
    VRAM, temperature, clock, and power of discrete cards) from sysfs, named
    from the driver or the PCI ID database; unknown values show as --; the
    GUI's GPU panel switches to shorter rows when several GPUs do not fit
  • sudo rigtop --install-collector sets up the root timer that collects
    SMART data and btrfs scrub results (the files of contrib/collect/ are
    built in, so cargo install is enough); --uninstall-collector removes it

Changed

  • btrfs: errors are reported per device name (nvme0n1: 3) instead of
    btrfs device id, looked up with the BTRFS_IOC_DEV_INFO ioctl (no root
    needed); alerts name the affected drives too

rigtop v0.5.1

Choose a tag to compare

@github-actions github-actions released this 25 Oct 15:51

Fixed

  • Overview: Linux hosts listed fewer processes than configured, because
    busy kernel threads took places in the top list the agent sends and
    were then hidden; the agent now leaves them out
  • Overview: the process list (and the GUI's swap row) keeps its height
    when there are fewer entries, so the sections below no longer jump

rigtop v0.5.0

Choose a tag to compare

@github-actions github-actions released this 25 Oct 14:55

Added

  • Laptops (Linux): battery charge and charge/discharge power in the Power
    line (Bat 72% -12W) and on overview cards; batteries of mice and
    keyboards are left out
  • AMD APUs (Linux): without readable RAPL, the APU's SoC power (amdgpu
    PPT) is shown as SoC and counted in the total
  • README: a Linux section with the packages per distribution (Debian /
    Ubuntu and Arch) and a udev rule that lets the admin group read RAPL

Changed

  • Discrete AMD GPUs' power is no longer reported as integrated GPU power
  • Linux: collecting processes no longer refreshes every thread, which cut
    the agent's CPU use to about a quarter on a desktop with thousands of
    browser threads; thread counts now come from /proc/<pid>/stat
  • Remote hosts: when rigtop is not on the PATH of the SSH session,
    the default agent command falls back to ~/.cargo/bin/rigtop --agent

rigtop v0.4.0

Choose a tag to compare

@github-actions github-actions released this 25 Oct 12:44

Added

  • Windows: temperatures, fans and CPU package power from
    LibreHardwareMonitor's web server (data.json, default
    http://127.0.0.1:8085; [lhm] url in the configuration), with a
    note in the CPU panel while it cannot be read
  • Windows GUI: "set up…" installs LibreHardwareMonitor with winget when
    needed, turns its web server on for this computer only and starts it