Skip to content

The TUI

Gopher Glide features a highly polished, interactive Terminal UI designed for real-time observability. Our latest revamp introduces a premium aesthetic powered by a semantic color palette, delivering a clean, modern experience directly in your terminal.

Gopher Glide TUI Demo

The dashboard provides an at-a-glance view of your load test’s vital signs, including configuration parameters, real-time throughput metrics, latency percentiles, and a visual stage timeline representing your current progress within the load profile.

The TUI is built on a custom styling engine. It supports full semantic theming so the entire dashboard can be cleanly restyled without touching component code.

Currently, Gopher Glide defaults to a beautifully balanced Catppuccin Macchiato inspired palette, ensuring your load tests look as good as they run.

The TUI is highly optimized to ensure it doesn’t interfere with the load test itself. It targets a smooth ~24 FPS (refreshing every 42ms) to provide a highly responsive, real-time observability experience without consuming excessive CPU cycles.

KeyAction
↑Increase live RPS by +5 (Director Mode bias)
↓Decrease live RPS by -5 (Director Mode bias)
fToggle call log between all calls and errors only
q / Ctrl+CQuit

While a run is in progress, use ↑ / ↓ to apply a live RPS bias on top of the configured stage target. The bias accumulates (e.g., three ↑ presses = +15 RPS) and is shown in the TUI and reflected in the live metrics. The bias is applied on top of the LERP’d value — the stage plan continues to run unaffected.

  • Where to look: a brief ▲ +5 RPS / ▼ -5 RPS flashes in the director bar when you press a key, and a BIAS +N RPS badge stays there while the total is non-zero.
  • When it lands: the rate changes at the next one-second window, and never drops below 1 RPS. The keys only act while the test is running, and only in the terminal that’s running the TUI.
  • It’s recorded: every press is saved in the snapshot’s bias_events (see Manual bias & marks), so a later gg snap diff can tell a biased run from a clean one.

Running headless — in CI, from a script, or from another process? The same bias, plus mark and stop, is available over stdin; see Driving a Run Live.