Everything behind this text is rendered on a terminal.
Ghostty's terminal engine and a Rust WebGPU renderer, compiled into one Wasm module. Drop it in where xterm.js was, or run Ghostty itself in a worker.
One dense screen: 240×80 cells rewritten every update, with about 4,800 style changes. Lower is better. Bars show milliseconds of CPU for parsing plus redraw.
Medians from interleaved runs, 10 Oct 2026: Chrome 155, Apple M5 Max, hardware WebGPU, Wraith build c6a77bd, xterm 6.0 with WebGL addon 0.19. GPU-completed updates, not display frames. Native figures drive the engine directly, without the worker hop. Your machine will differ, so race it yourself below.
This runs the repository's own benchmark harness: real xterm.js 6.0, its official WebGL addon, and both Wraith paths, each fed the same bytes and fenced on the GPU.
Needs a WebGPU browser. Takes about 20 s.
Six steps, all off your page's main thread, and nothing drawn twice.
Output from your shell, batched into one message per burst.
Everything after this runs off your page's main thread.
libghostty-vt reads every escape sequence the way Ghostty does.
Unchanged rows are skipped. Scrolled rows are reused, not redrawn.
Each cell is 20 bytes on the GPU: its colors, its style and one glyph.
Backgrounds, then glyphs. Empty cells cost nothing.
import { Terminal } from 'wraithterm/xterm';
import { FitAddon } from '@xterm/addon-fit';
const term = new Terminal({ fontSize: 14 });
term.loadAddon(new FitAddon());
term.open(element);
term.write('same API, WebGPU underneath');xterm's parser and buffer, Wraith's renderer. Addons such as fit, search, serialize and web links keep working.
import { Wraith } from 'wraithterm';
const term = await Wraith.mount(element);
term.connect(new WebSocket(ptyUrl));
term.onResize(({ cols, rows }) => {
// tell the PTY
});Ghostty's terminal semantics, Kitty keyboard encoding and reflow, with the whole terminal off the main thread.
Two hands-on labs, both running the real thing in your browser.
