netspd 0.1.2

A beautiful, modern, minimalistic network speed testing terminal application
Documentation

netspd

A beautiful, modern network speed test for your terminal — with a hypercar tachometer.

CI Crates.io License: MIT Rust

netspd measures ping, jitter, download and upload against LibreSpeed-compatible servers and renders them live on a braille-canvas instrument cluster: a spring-loaded needle with an ignition sweep, a heat-gradient band that reddens toward the hatched redline, a session-peak ghost notch, and a latency sub-dial — in the spirit of tools like btop, gitui and lazygit.

Features

  • Hypercar tachometer — heat-gradient value band, hatched redline, spring-physics needle with afterglow, ignition sweep before each phase, peak ghost notch, latency sub-dial; twin instrument cluster on wide terminals
  • Live metrics — current / average / peak speed, transferred bytes, ETA, elapsed time
  • Latency analysis — multiple samples, outlier trimming, jitter, real ICMP packet loss (graceful fallback where ICMP is unavailable)
  • Four providers — LibreSpeed (default), Ookla speedtest.net, Netflix Fast.com, and custom for your own servers (self-hosted backends, LAN testing); switch with --provider or config
  • Smart server discovery — health-probes the server list, drops dead servers, auto-selects the nearest
  • Headless mode — --no-tui for scripts and cron, --json for machine-readable reports
  • Result history — every run appended as JSON lines to your data directory
  • Streaming transfers — nothing buffered in memory; parallel connections with EMA smoothing
  • Themes — Default, Nord, Dracula, Catppuccin, Gruvbox, plus your own TOML themes without recompiling
  • Responsive layout — adapts from 80×24 up to 4K terminals
  • Graceful everywhere — cancellation, retries, timeouts; network failures never panic

Installation

Cargo

cargo install netspd

Compiles from crates.io; requires Rust 1.88 or newer.

Homebrew (macOS / Linux)

brew tap TarunVishwakarma1/tap
brew install netspd

Prebuilt binaries

Download the archive for your platform from the latest release, unpack it and put netspd on your PATH.

From source

git clone https://github.com/TarunVishwakarma1/netspd
cd netspd
cargo install --path .

Requires Rust 1.88 or newer.

Usage

netspd                     # full TUI
netspd --no-tui            # headless: progress on stderr, summary on stdout
netspd --json              # headless: report as one JSON object on stdout
netspd --list-servers      # print reachable servers, nearest first
netspd -s tokyo            # pick a server by name/host substring
netspd -p ookla            # provider: librespeed (default) | ookla | fast | custom
netspd -d 5 -c 8           # 5-second phases over 8 connections

The test starts automatically: ping → download → upload, then a results summary. If a server fails mid-test, netspd automatically retries with the next-nearest one (unless you pinned a server with --server).

Keyboard

Key Action
q / Esc Quit (Esc closes overlays first)
r Restart the test
g Result trends from your history
s Server selection
t Theme selector
c View configuration
? Help
↑↓ / jk, Enter Navigate and confirm in lists

Scripting

--json prints exactly one JSON object on stdout and nothing else, so it pipes cleanly:

netspd --json | jq .download_mbps
{
  "timestamp": 1783092092,
  "server": "Tokyo, Japan (A573)",
  "ping_ms": 141.2,
  "jitter_ms": 1.0,
  "packet_loss_pct": 0.0,
  "download_mbps": 93.7,
  "download_peak_mbps": 171.0,
  "download_bytes": 117193129,
  "upload_mbps": 66.3,
  "upload_peak_mbps": 87.8,
  "upload_bytes": 82837504
}

Exit code is 0 on a completed test and 1 on any failure. Every completed run (TUI or headless) is also appended to <data dir>/netspd/history.jsonl (macOS: ~/Library/Application Support/netspd/, Linux: ~/.local/share/netspd/), one JSON object per line.

Configuration

netspd reads the first config.toml found in:

  1. $XDG_CONFIG_HOME/netspd/config.toml (macOS: ~/Library/Application Support/netspd/config.toml)
  2. ./config/config.toml

Every key is optional; see config/config.toml for the annotated reference. Highlights:

theme = "catppuccin"       # default | nord | dracula | catppuccin | gruvbox
refresh_rate = 30          # UI frames per second (1..=60)
provider = "librespeed"    # librespeed | ookla | fast
animation_speed = 1.0

[engine]
duration_secs = 10         # length of each transfer phase
connections = 4            # parallel connections
ping_samples = 10

[[servers]]                # optional: pin your own backend
name = "My Server"
url = "https://speedtest.example.com/backend/"

Custom themes go in ~/.config/netspd/themes/*.toml — copy any file from assets/themes/ as a starting point.

Your own servers and LAN testing

provider = "custom" tests against exactly the servers you declare — a self-hosted backend, an office box, or a machine on your LAN. The endpoints just need to speak plain HTTP:

  • ping — answers GET with any 2xx, quickly
  • download — answers GET with a large body (netspd streams and loops requests)
  • upload — accepts POST bodies with a 2xx

A stock LibreSpeed container on the target machine provides all three:

docker run -d -p 8080:80 ghcr.io/librespeed/speedtest   # on the LAN machine
provider = "custom"

[[servers]]
name = "Office LAN"
url = "http://192.168.1.50:8080/backend/"
# paths default to LibreSpeed's (garbage.php / empty.php); override for
# other servers:
# download_path = "big-file.bin"
# upload_path   = "upload-sink"
# ping_path     = "health"

Then netspd -p custom. Loopback sanity check: this setup measures ~11 Gbps against a local process, so the meter itself won't be your bottleneck on 10GbE.

Architecture

netspd follows clean architecture with strict one-way dependencies:

┌──────────────────────────────────────────────┐
│ tui/          presentation (Ratatui only)    │
│   theme · animation · widgets · screens      │
├──────────────────────────────────────────────┤
│ app/          application                    │
│   events → controller → state → renderer    │
├──────────────────────────────────────────────┤
│ engine/       domain (no UI imports)         │
│   Engine → scheduler → providers/network     │
│   metrics: EMA · sampler · throughput ·      │
│            latency · statistics              │
├──────────────────────────────────────────────┤
│ config/ · errors/ · utils/   infrastructure  │
└──────────────────────────────────────────────┘
  • The engine emits strongly-typed EngineEvents over a channel and never imports Ratatui — it is directly reusable from a CLI, GUI, REST API or as a library.
  • Providers implement one trait (Provider); adding a new speed test network touches nothing else. LibreSpeed, Ookla and Fast.com are each ~100 lines of server discovery — the transfer engine is shared.
  • Note: Ookla runs over the servers' long-standing HTTP endpoints (the classic speedtest-cli protocol), not Ookla's newer socket protocol — upload figures may read lower than the official client on high-latency links.
  • State is plain data mutated only by reducers; the renderer just draws it.
  • Metrics are pure, deterministic calculators, each independently unit-tested.

Development

cargo test            # unit + integration tests
cargo clippy          # lint-clean, unwrap/expect/panic denied
cargo fmt --check

See CONTRIBUTING.md for the layering rules and PR checklist.

Roadmap

  • JSON result export and --no-tui CLI mode
  • Result history (JSON lines)
  • Trend sparklines over the stored history (g)
  • --server, --list-servers, --duration, --connections flags
  • Automatic failover to the next server on failure
  • Packet loss via ICMP echo probing
  • Homebrew tap
  • Ookla and Fast.com providers
  • Scheduled repeat testing

License

MIT