rigtop

Early development. The TUI and GUI work on Linux and Windows and monitor several hosts over SSH with an overview.

A btop-inspired system monitor with two front-ends that share one data-collection core:

• rigtop — terminal UI
• rigtop-gui — native GUI (no web view) with a btop-like layout and resizable panels

日本語版 README

Goals

• The btop overview on one screen: CPU (total, per core, temperatures), every GPU, memory, storage, network, and a process list with kill / force kill.
• Storage health, which existing monitors mostly skip: btrfs RAID profiles and per-device error counters, mdadm array state, NVMe temperatures, and SMART health.
• Per-process GPU usage and VRAM (NVIDIA, via NVML).
• Combined CPU + GPU power draw and fan speeds (with your own labels for fan channels).
• Multiple hosts over plain SSH: rigtop --agent streams snapshots over stdin/stdout, so there is no daemon to run and no port to open.

Roadmap

1. Collection core and the TUI on Linux (v0.1)
2. Native GUI on Linux (v0.1)
3. Windows support (v0.1; temperatures, fans and CPU power through LibreHardwareMonitor since v0.4; macOS builds without hardware testing)
4. Remote monitoring over SSH: one host (v0.2), several hosts with an overview (v0.3)

Installation

Download an archive for your platform from the Releases page and put the binaries on your PATH:

SHA256SUMS lists the checksums. With a Rust toolchain (1.95 or newer) you can also install from crates.io:

cargo install rigtop rigtop-gui

On Arch Linux and derivatives, packaging/aur/rigtop/PKGBUILD builds a package (it will be on the AUR once account registration there reopens):

git clone https://github.com/akivajp/rigtop.git
cd rigtop/packaging/aur/rigtop && makepkg -si

Usage

cargo run --release -p rigtop            # terminal UI
cargo run --release -p rigtop -- --help  # options (--interval-ms, --config, --dump-json)

GUI

cargo run --release -p rigtop-gui            # same options: --interval-ms, --config

Drag panel borders to resize and panel titles to rearrange. Click a column header to sort, right-click a process to terminate or kill it (asks first), and pick the network interface from the list.

Zoom with Ctrl + mouse wheel, Ctrl + + / - / 0, or the A- / A+ buttons in the CPU panel title. The zoom level and panel layout are restored on the next start; "reset layout" brings back the default arrangement.

When a new alert appears on any host (drive errors, a missing RAID device, failing SMART, a nearly full filesystem), the GUI shows a desktop notification, also while it is minimized (Linux and macOS; not on Windows yet). Alerts present at startup are not announced. [notifications] level in the configuration picks "critical" (default), "warning" (also resyncs, filling filesystems, hot GPUs) or "off"; rigtop-gui --test-notification shows a sample notification.

With mouse support on, most terminals select text with Shift+drag.

Remote monitoring over SSH

Install rigtop on the remote hosts too, then list them in the configuration file or on the command line:

# ~/.config/rigtop/config.toml
[[hosts]]
name = "zefat"                 # also the ssh destination unless `ssh` is given

[[hosts]]
name = "nas"
ssh = "admin@nas.local"
command = "/usr/local/bin/rigtop --agent"
sections = ["storage", "cpu", "network"]   # what this host's overview card shows first

rigtop                          # TUI: hosts from the configuration
rigtop-gui --host hermon        # GUI: plus a host from the command line (repeatable)

In the GUI, hosts… (in the tab bar, or in the CPU panel title while only this machine is shown) adds, edits and removes hosts and can test the connection first. Changes are written to the [[hosts]] of the configuration file, so the TUI sees them too; the rest of the file is left as it is.

With more than one host, rigtop opens on an overview with one card per host, plus a tab per host (this machine included) for the full detail view. Cards show alerts (storage errors, failing SMART, degraded RAID, nearly full filesystems, hot GPUs) and then CPU, GPU, memory, the top processes, network and storage, in that order and as far as the card has room. Change the default order with [overview] sections, per host with sections, or in the GUI with each card's "sections" menu. Whether this machine gets a card is [overview] include_local (or the checkbox in the GUI, l in the TUI). In the TUI, Tab and 0-9 switch between the overview and the hosts.

rigtop starts rigtop --agent on each remote host through your system's ssh (the built-in OpenSSH client on Windows), so ~/.ssh/config, keys and ProxyJump work as usual. There is no daemon to run and no port to open; the agent exits when the connection closes, and rigtop reconnects by itself. Hosts not shown in detail only send their top processes.

• Authentication must work without a prompt (key or agent), and the host key must already be known: connect once with ssh <host> first.
• Non-interactive SSH sessions often lack ~/.cargo/bin on their PATH (zsh reads only ~/.zshenv for them, bash usually nothing). When rigtop is not found there, rigtop tries ~/.cargo/bin/rigtop --agent by itself. For other locations, set command for the host (or --remote-command for --host).
• Killing processes on a remote host uses the SSH user's permissions.

Linux

Most readings need neither extra software nor privileges: CPU, memory, processes, temperatures and fans (hwmon), network, disks, btrfs and md RAID, and laptop batteries. A few need a package or a permission:

CPU power (RAPL)

Since Linux 5.10 the CPU energy counters (/sys/class/powercap/intel-rapl:*/energy_uj, named so on AMD too) are readable only by root, because fine-grained power readings can leak secrets (PLATYPUS, CVE-2020-8694). A udev rule can let your admin group read them; replace ADMIN_GROUP with sudo on Debian/Ubuntu or wheel on Arch:

echo 'SUBSYSTEM=="powercap", KERNEL=="intel-rapl:*", RUN+="/usr/bin/chgrp ADMIN_GROUP /sys%p/energy_uj", RUN+="/usr/bin/chmod g+r /sys%p/energy_uj"' \
  | sudo tee /etc/udev/rules.d/90-rigtop-rapl.rules
sudo udevadm control --reload && sudo udevadm trigger --subsystem-match=powercap

Without it, AMD laptops show the APU's SoC power (SoC) instead, and other machines show --.

Windows

Both rigtop and rigtop-gui run on Windows (see Installation). Builds of the latest main are also attached to the CI runs (artifact rigtop-windows-x86_64).

What differs from Linux:

• CPU load, memory, processes, network (with link speeds), drive capacity and per-drive I/O, and NVIDIA GPUs (via nvml.dll from the driver) work without extra software.
• Temperatures, fans, CPU power and drive health (wear, spare capacity, temperature) need hardware drivers that Windows does not ship. rigtop reads them from LibreHardwareMonitor when it is running (see below); otherwise only what Windows itself reports is shown, which is often nothing. Drives are named disk0, disk1... as in Disk Management.
• There is no load average and no disk busy percentage.
• Windows has no signals: "kill" (K) ends the process immediately; "terminate" (k) asks the process's windows to close, like Task Manager's "End task", and reports when the process has no window (console programs, services).

Temperatures, fans and CPU power with LibreHardwareMonitor

In rigtop-gui, set up… next to the note in the CPU panel does steps 1 to 3 for you: it installs LibreHardwareMonitor with winget if needed, turns its web server on for this computer only (127.0.0.1) and starts it. By hand:

1. Install it, together with the PawnIO driver it needs for CPU sensors (Windows asks for administrator approval for the driver):

   winget install LibreHardwareMonitor.LibreHardwareMonitor
   

   (Or download LibreHardwareMonitor.zip from its releases, unblock it in the file's Properties, and extract it.)
2. Run LibreHardwareMonitor from a new terminal (or the extracted LibreHardwareMonitor.exe) and allow it to run as administrator.
3. Turn on Options → Remote Web Server → Run (port 8085). rigtop reads http://127.0.0.1:8085/data.json and picks the values up within 10 seconds; until then, the CPU panel shows a note.
4. Optionally enable Options → Run On Windows Startup and Start Minimized.

The web server must not require authentication. Its port can be reached from other machines unless Windows Firewall blocks it. For another address or port, set [lhm] url in the configuration; url = "" turns this off. NVIDIA GPUs keep coming from the driver, not LibreHardwareMonitor.

SMART and btrfs scrub results (optional)

SMART data and btrfs scrub results are readable only by root, and rigtop never runs as root or loosens device permissions. Instead, a root systemd timer collects them every 10 minutes into world-readable files under /var/lib/rigtop/, and rigtop reads those:

• smart/<device>.json: smartctl -j -a for every drive (sleeping disks are not woken up)
• btrfs/scrub.status.<fsid>: copies of btrfs-progs' scrub status, shown as when data was last verified

sudo apt install smartmontools   # Arch: sudo pacman -S smartmontools
sudo "$(command -v rigtop)" --install-collector

--install-collector writes the script and the systemd units (the files in contrib/collect/), starts the timer and collects once right away; --uninstall-collector removes them and the collected files. The full path is needed because sudo usually does not search ~/.cargo/bin. From a checkout, the same by hand:

sudo install -m 755 contrib/collect/rigtop-collect /usr/local/libexec/
sudo install -m 644 contrib/collect/rigtop-collect.service contrib/collect/rigtop-collect.timer /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now rigtop-collect.timer

The SMART files include drive serial numbers.

Configuration

~/.config/rigtop/config.toml (%APPDATA%\rigtop\config.toml on Windows). Every setting is optional.

[sensors]
# Names for sensor channels, keyed by the channel key rigtop shows.
labels = { "nct6686:nct6683.2592/fan1" = "CPU fan" }
# Channels to hide, e.g. unconnected thermistors reading -40 °C.
hidden = ["nct6686:nct6683.2592/temp4"]

[smart]
dir = "/var/lib/rigtop/smart"

[btrfs]
scrub_dir = "/var/lib/rigtop/btrfs"

[notifications]
# rigtop-gui: desktop notifications for new alerts ("critical", "warning" or "off").
level = "critical"

[lhm]
# Windows: LibreHardwareMonitor's web server ("" to disable).
url = "http://127.0.0.1:8085/data.json"

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option.