cosmotop is a system monitoring tool distributed as a single executable for multiple platforms. A fork of btop++ and built with Cosmopolitan Libc.

Installation

Download cosmotop from GitHub releases. Place it anywhere and run!

Homebrew

The Homebrew tap bjia56/tap supports installing the latest cosmotop from GitHub releases on both MacOS and Linux.

brew tap bjia56/tap
brew install cosmotop

Docker

A Docker image is available for Linux x86_64 and aarch64 hosts.

docker run -it --rm --net=host --pid=host ghcr.io/bjia56/cosmotop:latest

Windows setup notes

On Windows, rename cosmotop to either cosmotop.cmd or cosmotop.bat before running. This allows Windows to execute the file as a batch script, which can then properly self-extract and execute the embedded executable.

Usage and features

Usage: cosmotop [OPTIONS]

Options:
  -h,  --help          show this help message and exit
  -v,  --version       show version info and exit
  -lc, --low-color     disable truecolor, converts 24-bit colors to 256-color
  -t,  --tty_on        force (ON) tty mode, max 16 colors and tty friendly graph symbols
  +t,  --tty_off       force (OFF) tty mode
  -p,  --preset <id>   start with preset, integer value between 0-9
  -u,  --update <ms>   set the program update rate in milliseconds
  -o,  --option        override a configuration option in KEY=VALUE format, can use multiple times
       --show-defaults print default configuration values to stdout
       --show-themes   list all available themes
       --licenses      display licenses of open-source software used in cosmotop
       --debug         start in DEBUG mode: shows microsecond timer for information collect
                       and screen draw functions and sets loglevel to DEBUG
       --mcp           start MCP server mode: exposes system information tools via MCP protocol

GPU monitoring

Monitoring of GPUs is supported on Linux, MacOS, Windows.

• Windows: LibreHardwareMonitor is included with cosmotop and automatically used to fetch GPU information.
• Linux: Intel, AMD, and NVIDIA GPUs are supported, provided the appropriate driver is installed, and the following:
  • Intel: Root privileges are required to access metrics directly. Alternatively, run intel-gpu-exporter in a privileged Docker container, then use the Prometheus configuration options below.
  • AMD: rocm_smi_lib is statically linked and should work out of the box.
  • NVIDIA: libnvidia-ml.so must be available.
• MacOS: Utilization monitoring of the GPU is supported on Apple Silicon. Sudo is not required.

NPU monitoring

Monitoring of NPUs is supported on Linux and MacOS.

• Linux: Utilization monitoring of Intel and Rockchip NPUs is supported on Linux, provided the following:
  • Intel: The path /sys/devices/pci0000:00/0000:00:0b.0 must be readable.
  • Rockchip: The path /sys/kernel/debug/rknpu must be readable.
• MacOS: Utilization monitoring of the Apple Neural Engine is supported on Apple Silicon. Sudo is not required.

GPU and NPU monitoring via Prometheus

cosmotop supports polling an HTTP Prometheus endpoint to read GPU and NPU metrics data. The prometheus_endpoint configuration key should be set to the HTTP endpoint which publishes the metrics.

Some options for exporters and their recommended cosmotop configuration values:

• Linux: intel-gpu-exporter

  prometheus_mapping = "gpu_utilization_percent:igpu_engines_busy_max,gpu_frequency:igpu_frequency_actual,gpu_power_usage:igpu_power_gpu"
  prometheus_settings = "gpu_frequency_unit:mhz,gpu_power_unit:w"
  

• MacOS: [macmon-prometheus-exporter](https://github.com/D

... View full README on GitHub