A compact, single-file system information collector for Linux, macOS, BSD, and Solaris.
The default mode outputs a single line of machine-readable system telemetry, colored with ANSI 256 / truecolor themes. --all, --all-safe, and --drivers render multi-line tables for when you want the full picture. Zero external dependencies — just Python 3.8+ and a Unix-like operating system.
Host: atlas - OS: Linux 6.8.0/x86_64 - Distro: Ubuntu 24.04 - CPU: 8 x AMD Ryzen 7 5800X (3700.00 MHz) - GPU: NVIDIA GeForce RTX 3070 8192MB (5888 CUDA cores, 184 tensor cores (3rd gen), CC 8.6, SM 1725MHz/mem 7000MHz, 220W, PCIe 4x16, drv 535.183.01) - Processes: 312 - Uptime: 14d 6h 32m - Users: 1 - Load Average: 0.83 - Memory Usage: 7842.50MB/16384.00MB (47.86%) - Disk Usage: 186.42GB/512.00GB (36.41%)
Author: Jeremy Anderson
Website: git.dcos.net/dcosnet/probefetch
License: MIT
- Default mode — hostname, OS, distro, CPU, GPU, processes, uptime, users, load average, memory, disk, network traffic.
- Seven special modes —
--devel,--admin,--devops,--kernel,--pkgs,--security,--net— each outputs its own compact line. - Driver matrix —
--driversenumerates every PCI and USB device, marks each as bound or missing a driver, and resolves the kernel module(s) that claim it frommodules.alias. Built for distro developers assembling minimal driver sets. - Full report —
--allruns every default section and every special mode and renders the results as a single clean two-column table. - Safe report —
--all-safeis--allwith machine-identifying fields redacted (hostname, MAC NIC-unique half, IPv4 address, gateway, DNS) for pasting into forums, public screenshots, or shared images. The hardware profile is left intact. - Stealth / sleuth mode —
--sleuthor--stealthavoids spawning any subprocesses. Only file-based probes (/proc,/sys,/etc) are used, making it safe for constrained or audited environments. - Color themes — auto-detected terminal color depth, with eight built-in themes:
auto,dark,light,solarized,dracula,gruvbox,nord,mono. - Per-section toggles — show or hide any section by name, or use
--batteryand--networkto include those by default. - 60+ distro detection — identifies Linux distributions through
/etc/os-release,/usr/lib/os-release, release-file fingerprints, and/etc/issuefallback. - Cross-architecture CPU detection — handles x86, ARM, Alpha, IA-64, MIPS, PA-RISC, PowerPC, S/390, SH, and SPARC.
- GPU detection — NVIDIA (via
nvidia-smi): model, VRAM, CUDA cores, tensor cores, compute capability, driver, max clocks, power, and PCIe link; AMD (viarocminfo): model, VRAM, compute units, SIMDs/CU, peak clock; and anlspci+ sysfs PCI fallback for any vendor. - Accurate memory metrics — uses kernel
MemAvailable(since 3.14) when present, falling back to the traditionalMemFree + Buffers + Cachedestimation. - Filtered disk metrics — on Linux, aggregates only physical block devices (
/dev/sd*,/dev/nvme*,/dev/vd*, etc.), excluding loop devices, network mounts, and virtual filesystems.
| Flag | What it shows |
|---|---|
--devel |
Languages, compilers, build tools, and package managers (Python, Node, Go, Rust, Java, PHP, Ruby, Perl, GCC, CMake, Git, pip, npm, Cargo, etc.) |
--admin |
Administration panels, databases, monitoring, and web servers (Cockpit, MariaDB, PostgreSQL, Prometheus, Grafana, nginx, Apache, Caddy, Redis, RabbitMQ, etc.) |
--devops |
Containers, orchestration, IaC, CI/CD, and cloud CLIs (Docker, Podman, kubectl, Helm, OpenTofu, Ansible, Jenkins, AWS CLI, GitHub CLI, etc.) |
--kernel |
Kernel version, compiler, security modules (SELinux, AppArmor, Smack, Yama), and loaded module count |
--pkgs |
Installed package count per detected package manager (dpkg, rpm, pacman, apk, emerge, xbps, nix, dnf, zypper, sorcery) |
--security |
Firewall status, hardening tools (UFW, firewalld, Fail2Ban, Lynis, ClamAV, etc.), ASLR, dmesg/kptr restrictions |
--net |
Network interfaces, IPs, gateway, DNS servers, and TCP connection counts |
--drivers |
PCI/USB device matrix: bound drivers, missing drivers, and hardware-to-module matching resolved from modules.alias (for distro developers). |
--all |
Every default section and every special mode rendered as a single clean two-column table. |
--all-safe |
Same as --all but with machine-identifying fields redacted (hostname, MAC NIC-unique half, IPv4 address, gateway, DNS) for forum / public-screenshot pasting. |
--drivers walks /sys/bus/pci/devices and /sys/bus/usb/devices and produces a
multi-line report aimed at distro developers who need to know exactly which kernel
modules a target machine requires. For every device it records:
- the PCI/USB address and a human-readable class label,
- the vendor and device IDs (with a friendly vendor name where known),
- the driver currently bound and the kernel module behind it,
- the candidate modules that claim the device's modalias, resolved from
/lib/modules/<kernel>/modules.alias, - a status of BOUND (a driver is attached) or MISSING (nothing is bound).
A summary line reports totals, and a "Missing drivers" section lists each unbound
device alongside the module(s) that modules.alias suggests — the
hardware-to-module matching a distro developer needs to ship a minimal but complete
driver set.
$ python3 probefetch.py --drivers
PCI devices (4 found, 2 bound, 2 missing)
Address Class Vendor:Device Driver Module Status Candidates
------------ ----------- --------------------- ------- ------- ------- ----------
0000:00:00.0 Host Bridge Intel 0x8086:0x5904 - - MISSING intel_gmc
0000:00:02.0 VGA Display Intel 0x8086:0x5916 i915 i915 BOUND i915
0000:00:1f.6 Ethernet Realtek 0x10ec:0x8168 - - MISSING r8169
0000:01:00.0 3D Display NVIDIA 0x10de:0x1b81 nouveau nouveau BOUND nouveau,nvidia
Missing drivers - hardware to module matching:
PCI 0000:00:00.0 Host Bridge (0x8086:0x5904) -> intel_gmc
PCI 0000:00:1f.6 Ethernet (0x10ec:0x8168) -> r8169
Summary: 4 devices, 2 drivers bound, 2 missing
modules.alias: /lib/modules/6.8.0-40-generic/modules.alias (18234 aliases)
Known names (8 distinct):
Bound drivers (3): i915, nouveau, usbhid
Bound modules (3): i915, nouveau, usbhid
Candidates (8): i915, intel_gmc, nouveau, nvidia, r8169, uas, usb-storage, usbhid
Name references (kernel tree /usr/src/linux):
Module Description Source
----------- ---------------------------------------------- ------
i915 Intel Graphics tree
intel_gmc (no description found) -
nouveau nVidia Riva/TNT/GeForce/Quadro/Tesla/Tegra K1+ tree
nvidia (no description found) -
r8169 RealTek RTL-8169 Gigabit Ethernet driver tree
uas USB Attached SCSI driver tree
usb-storage USB Mass Storage driver for Linux tree
usbhid USB HID core driver tree
"MISSING" means no driver is currently bound — which is expected for some
classes (host bridges, ISA bridges) and a real gap for others (an Ethernet
controller with no r8169 loaded). Pair the candidate list with
modprobe --dry-run or your initramfs generator to decide what to ship.
The report closes with a Known names summary — the deduplicated set of
driver and module names observed (bound drivers, bound modules, and candidate
modules) — followed by a Name references table that resolves each name to
the MODULE_DESCRIPTION string from a Linux kernel source tree. This turns a
bare r8169 into "RealTek RTL-8169 Gigabit Ethernet driver" so the
human-readable purpose of every module is visible at a glance. The tree is
located in this order: the PROBEFETCH_KERNEL_TREE environment variable,
/usr/src/linux, then the newest /usr/src/linux-* directory; if a kernel
tree is available but modinfo is not (or --sleuth is set), descriptions are
resolved purely by reading the source (Makefile/Kbuild object lists and
MODULE_DESCRIPTION / #define macros) with no subprocesses. When neither a
tree nor modinfo is available the names still list, with descriptions marked
"(no description found)". To point probefetch at an extracted source tree
without root access, for example the 7.3.4 release:
$ export PROBEFETCH_KERNEL_TREE=/path/to/linux-7.3.4
$ python3 probefetch.py --drivers
--all enables every default section (including battery and network traffic) and
runs every special mode, then lays the results out as one bordered two-column
table — the closest thing probefetch has to a "show me everything" view.
$ python3 probefetch.py --all
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
| Category | Value |
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
| Host | atlas |
| OS | Linux 6.8.0-40-generic/x86_64 |
| Distro | Ubuntu 24.04 |
| CPU | 8 x AMD Ryzen 7 5800X (3700.00 MHz) |
| GPU | NVIDIA GeForce RTX 3070 8192MB (5888 CUDA cores, 184 tensor cores (3rd gen), CC 8.6, SM 1725MHz/mem 7000MHz, 220W, PCIe 4x16, drv 535.183.01) |
| Processes | 312 |
| Uptime | 14d 6h 32m |
| Users | 1 |
| Load Average | 0.83 |
| Memory Usage | 7842.50MB/16384.00MB (47.86%) |
| Disk Usage | 186.42GB/512.00GB (36.41%) |
| Kernel | 6.8.0-40-generic (gcc 13.2.0) - Security: AppArmor - Modules: 284 |
| Packages | dpkg: 2847 |
| Development | Python 3.12.3 - Node.js 20.11.0 - Rust 1.76.0 - GCC 13.2.0 ... |
| Admin | PostgreSQL, nginx, Redis |
| DevOps | Docker, kubectl, Helm, Ansible |
| Security | Tools: UFW, Fail2Ban - UFW: active - ASLR: full |
| Network | IF: eth0 (aa:bb:cc:dd:ee:ff) - IP: 192.168.1.42 - GW: 192.168.1.1 |
| Drivers | 12 devices, 10 bound, 2 missing - missing needs: r8169 |
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
--all respects --theme=NAME and --sleuth like every other mode.
probefetch resolves the GPU(s) and reports a compact but vector-rich string per card. The strategy depends on what is available on the host:
-
NVIDIA (
nvidia-smi) — one query pulls the model, VRAM, CUDA cores, compute capability, driver version, max SM and memory clocks, default power limit, and PCIe link generation × width. The tensor-core count is derived from the CUDA-core count using the documented per-SM ratio for each compute capability (nvidia-smi does not report it directly); the tensor-core generation is stated alongside it.GPU: NVIDIA GeForce RTX 3070 8192MB (5888 CUDA cores, 184 tensor cores (3rd gen), CC 8.6, SM 1725MHz/mem 7000MHz, 220W, PCIe 4x16, drv 535.183.01) -
AMD (
rocminfo) — when the ROCm runtime is installed, the marketing name, VRAM, compute-unit count, SIMDs per CU, and peak clock are read fromrocminfo. CPU agents are skipped.GPU: AMD Radeon RX 6800 XT 16384MB (60 CUs, 2 SIMD/CU, 2500 MHz) -
sysfs +
lspcifallback — for any vendor (AMD, Intel, ASPEED, …) withoutnvidia-smi/rocminfo, the PCI class codes in/sys/bus/pci/deviceslocate the GPU,lspci+pci.idssupply the model name, andmem_info_vram_totalsupplies VRAM. Withoutlspci, the vendor and PCI device id are shown (NVIDIA [0x2484] (8192MB VRAM)).
The tensor-core derivation uses documented per-SM ratios (CUDA cores per SM and
tensor cores per SM) for compute capabilities 7.0, 7.5, 8.0, 8.6, 8.7, 8.9, and
9.0; for newer or undocumented capabilities the generation is reported without a
fabricated count, and pre-Volta hardware (CC < 7.0) reports "no tensor cores".
Missing or N/A fields from older drivers are dropped silently, so the line
degrades gracefully rather than failing.
--all-safe produces exactly the same two-column table as --all, but redacts
the fields that uniquely identify a machine — so the output is safe to paste into
forum signatures, public screenshots, or shared images without leaking the box's
network identity. The hardware profile is left untouched.
| Field | --all |
--all-safe |
|---|---|---|
| Hostname | real short hostname | redacted |
| MAC address | full MAC | OUI kept, NIC half masked → 06:90:4e:XX:XX:XX |
| IPv4 address | real address | redacted |
| Gateway | real address | redacted |
| DNS servers | real addresses | redacted (N servers) |
| Interface name | as-is | classic / PCI-slot names unchanged; MAC-derived enx<MAC> / wlx<MAC> masked |
| Everything else (CPU, GPU, RAM, disk, distro, kernel, packages, drivers, …) | unchanged | unchanged |
The MAC keeps its first three octets — the OUI, which identifies the vendor and is part of the hardware profile — and masks the last three, which uniquely identify the NIC. Connection counts (established / listening) and the interface class are preserved, since they are not unique to the machine.
$ python3 probefetch.py --all-safe
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
| Category | Value |
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
| Host | redacted |
| OS | Linux 6.8.0-40-generic/x86_64 |
| Distro | Ubuntu 24.04 |
| CPU | 8 x AMD Ryzen 7 5800X (3700.00 MHz) |
| GPU | NVIDIA GeForce RTX 3070 8192MB (5888 CUDA cores, 184 tensor cores (3rd gen), CC 8.6, SM 1725MHz/mem 7000MHz, 220W, PCIe 4x16, drv 535.183.01) |
| Processes | 312 |
| Uptime | 14d 6h 32m |
| Users | 1 |
| Load Average | 0.83 |
| Memory Usage | 7842.50MB/16384.00MB (47.86%) |
| Disk Usage | 186.42GB/512.00GB (36.41%) |
| Kernel | 6.8.0-40-generic (gcc 13.2.0) - Security: AppArmor - Modules: 284 |
| Packages | dpkg: 2847 |
| Development | Python 3.12.3 - Node.js 20.11.0 - Rust 1.76.0 - GCC 13.2.0 ... |
| Admin | PostgreSQL, nginx, Redis |
| DevOps | Docker, kubectl, Helm, Ansible |
| Security | Tools: UFW, Fail2Ban - UFW: active - ASLR: full |
| Network | IF: eth0 (06:90:4e:XX:XX:XX) - IP: redacted - GW: redacted - DNS: redacted (2 servers) ... |
| Drivers | 12 devices, 10 bound, 2 missing - missing needs: r8169 |
+--------------+-----------------------------------------------------------------------------------------------------------------------------------------------+
--all-safe combines with --theme=NAME and --sleuth. It is intentionally a
single self-contained report — the redaction applies only to the --all-safe
table, so --all and the standalone --net still show real values when you
need them.
--theme=auto # Auto-detect terminal color depth (default)
--theme=dark # Cornflower blue labels on dark terminals
--theme=light # Navy labels on light terminals
--theme=solarized # Solarized palette
--theme=dracula # Dracula palette
--theme=gruvbox # Gruvbox palette
--theme=nord # Nord palette
--theme=mono # Bold/dim only, no color
Themes adapt automatically to the terminal's color support: truecolor (24-bit) when available, xterm-256 as fallback, 16-color as last resort, and plain text when stdout is not a TTY.
| Platform | Default Mode | Special Modes | Stealth Mode |
|---|---|---|---|
| Linux | Full | Full | Full |
| macOS | Partial | Partial | Limited |
| FreeBSD | Partial | Partial | Limited |
| OpenBSD | Partial | Partial | Limited |
| NetBSD | Partial | Partial | Limited |
| DragonFly BSD | Partial | Partial | Limited |
| Solaris | Partial | Limited | Limited |
"Partial" means some information (e.g., GPU, battery) may be unavailable depending on the specific system configuration. Stealth mode limits all platforms to file-based probes only.
Show only the sections you care about by passing their names as arguments:
probefetch.py hostname cpu memory disk
probefetch.py distro uptime loadaverage
probefetch.py --battery --networkAvailable section names: hostname, os, distro, cpu, gpu, processes, uptime, users, loadaverage, battery, memory, disk, network.
When section names are given, all other sections are disabled.
- Python 3.8 or later
- A Unix-like operating system (Linux, macOS, BSD, Solaris)
- No external Python packages required
Based on sysinfo.pl by:
- David Rudie <d.rudie@gmail.com>
- Travis Morgan <imbezol@criticaldamage.com>
- Nils Goers <weechatter@arcor.de>
Rewritten in Python as probefetch by Jeremy Anderson with expanded platform support, color themes, stealth mode, and additional special modes.
MIT License. See LICENSE for the full text.